PyImageSearch

You can master Computer Vision, Deep Learning, and OpenCV - PyImageSearch

Deep Learning Object Detection OpenCV Tutorials Tutorials

Object detection with deep learning and OpenCV

by on

Last updated on July 7, 2021.

A couple weeks ago we learned how to classify images using deep learning and OpenCV 3.3’s deep neural network (dnn ) module.

While this original blog post demonstrated how we can categorize an image into one of ImageNet’s 1,000 separate class labels it could not tell us where an object resides in image.

In order to obtain the bounding box (x, y)-coordinates for an object in a image we need to instead apply object detection.

Object detection can not only tell us what is in an image but also where the object is as well.

Object detection algorithms need diverse and high-quality data to perform optimally. A rich dataset library helps train more accurate and adaptable models, ready for real-world detection tasks.

Roboflow has free tools for each stage of the computer vision pipeline that will streamline your workflows and supercharge your productivity.

Sign up or Log in to your Roboflow account to access state of the art dataset libaries and revolutionize your computer vision pipeline.

You can start by choosing your own datasets or using our PyimageSearch’s assorted library of useful datasets.

Bring data in any of 40+ formats to Roboflow, train using any state-of-the-art model architectures, deploy across multiple platforms (API, NVIDIA, browser, iOS, etc), and connect to applications or 3rd party tools.

With a few images, you can train a working computer vision model in an afternoon. For example, bring data into Roboflow from anywhere via API, label images with the cloud-hosted image annotation tool, kickoff a hosted model training with one-click, and deploy the model via a hosted API endpoint. This process can be executed in a code-centric way, in the cloud-based UI, or any mix of the two.

Over 250,000 developers and machine learning engineers from companies such as Cardinal Health, Walmart, USG, Rivian, Intel, and Medtronic build computer vision pipelines with Roboflow. Get started today, no credit card required.

In the remainder of today’s blog post we’ll discuss how to apply object detection using deep learning and OpenCV.

  • Update July 2021: Added a section on alternative deep learning-based object detectors, including articles on how to train R-CNNs from scratch, and more details on bounding box regression.

Looking for the source code to this post?

Jump Right To The Downloads Section

Object detection with deep learning and OpenCV

In the first part of today’s post on object detection using deep learning we’ll discuss Single Shot Detectors and MobileNets.

When combined together these methods can be used for super fast, real-time object detection on resource constrained devices (including the Raspberry Pi, smartphones, etc.)

From there we’ll discover how to use OpenCV’s dnn module to load a pre-trained object detection network.

This will enable us to pass input images through the network and obtain the output bounding box (x, y)-coordinates of each object in the image.

Finally we’ll look at the results of applying the MobileNet Single Shot Detector to example input images.

In a future blog post we’ll extend our script to work with real-time video streams as well.

Single Shot Detectors for object detection

Figure 1: Examples of object detection using Single Shot Detectors (SSD) from Liu et al.

When it comes to deep learning-based object detection there are three primary object detection methods that you’ll likely encounter:

Faster R-CNNs are likely the most “heard of” method for object detection using deep learning; however, the technique can be difficult to understand (especially for beginners in deep learning), hard to implement, and challenging to train.

Furthermore, even with the “faster” implementation R-CNNs (where the “R” stands for “Region Proposal”) the algorithm can be quite slow, on the order of 7 FPS.

If we are looking for pure speed then we tend to use YOLO as this algorithm is much faster, capable of processing 40-90 FPS on a Titan X GPU. The super fast variant of YOLO can even get up to 155 FPS.

The problem with YOLO is that it leaves much accuracy to be desired.

SSDs, originally developed by Google, are a balance between the two. The algorithm is more straightforward (and I would argue better explained in the original seminal paper) than Faster R-CNNs.

We can also enjoy a much faster FPS throughput than Ren et al. at 22-46 FPS depending on which variant of the network we use. SSDs also tend to be more accurate than YOLO. To learn more about SSDs, please refer to Liu et al.

MobileNets: Efficient (deep) neural networks

Figure 2: (Left) Standard convolutional layer with batch normalization and ReLU. (Right) Depthwise separable convolution with depthwise and pointwise layers followed by batch normalization and ReLU (figure and caption from Liu et al.).

When building object detection networks we normally use an existing network architecture, such as VGG or ResNet, and then use it inside the object detection pipeline. The problem is that these network architectures can be very large in the order of 200-500MB.

Network architectures such as these are unsuitable for resource constrained devices due to their sheer size and resulting number of computations.

Instead, we can use MobileNets (Howard et al., 2017), another paper by Google researchers. We call these networks “MobileNets” because they are designed for resource constrained devices such as your smartphone. MobileNets differ from traditional CNNs through the usage of depthwise separable convolution (Figure 2 above).

The general idea behind depthwise separable convolution is to split convolution into two stages:

  1. A 3×3 depthwise convolution.
  2. Followed by a 1×1 pointwise convolution.

This allows us to actually reduce the number of parameters in our network.

The problem is that we sacrifice accuracy — MobileNets are normally not as accurate as their larger big brothers…

…but they are much more resource efficient.

For more details on MobileNets please see Howard et al.

Combining MobileNets and Single Shot Detectors for fast, efficient deep-learning based object detection

If we combine both the MobileNet architecture and the Single Shot Detector (SSD) framework, we arrive at a fast, efficient deep learning-based method to object detection.

The model we’ll be using in this blog post is a Caffe version of the original TensorFlow implementation by Howard et al. and was trained by chuanqi305 (see GitHub).

The MobileNet SSD was first trained on the COCO dataset (Common Objects in Context) and was then fine-tuned on PASCAL VOC reaching 72.7% mAP (mean average precision).

We can therefore detect 20 objects in images (+1 for the background class), including airplanes, bicycles, birds, boats, bottles, buses, cars, cats, chairs, cows, dining tables, dogs, horses, motorbikes, people, potted plants, sheep, sofas, trains, and tv monitors.

Deep learning-based object detection with OpenCV

In this section we will use the MobileNet SSD + deep neural network (dnn ) module in OpenCV to build our object detector.

I would suggest using the “Downloads” code at the bottom of this blog post to download the source code + trained network + example images so you can test them on your machine.

Let’s go ahead and get started building our deep learning object detector using OpenCV.

Open up a new file, name it deep_learning_object_detection.py , and insert the following code:

# import the necessary packages
import numpy as np
import argparse
import cv2

# construct the argument parse and parse the arguments
ap = argparse.ArgumentParser()
ap.add_argument("-i", "--image", required=True,
	help="path to input image")
ap.add_argument("-p", "--prototxt", required=True,
	help="path to Caffe 'deploy' prototxt file")
ap.add_argument("-m", "--model", required=True,
	help="path to Caffe pre-trained model")
ap.add_argument("-c", "--confidence", type=float, default=0.2,
	help="minimum probability to filter weak detections")
args = vars(ap.parse_args())

On Lines 2-4 we import packages required for this script — the dnn module is included in cv2 , again, making hte assumption that you’re using OpenCV 3.3.

Then, we parse our command line arguments (Lines 7-16):

  • --image : The path to the input image.
  • --prototxt : The path to the Caffe prototxt file.
  • --model : The path to the pre-trained model.
  • --confidence : The minimum probability threshold to filter weak detections. The default is 20%.

Again, example files for the first three arguments are included in the “Downloads” section of this blog post. I urge you to start there while also supplying some query images of your own.

Next, let’s initialize class labels and bounding box colors:

# initialize the list of class labels MobileNet SSD was trained to
# detect, then generate a set of bounding box colors for each class
CLASSES = ["background", "aeroplane", "bicycle", "bird", "boat",
	"bottle", "bus", "car", "cat", "chair", "cow", "diningtable",
	"dog", "horse", "motorbike", "person", "pottedplant", "sheep",
	"sofa", "train", "tvmonitor"]
COLORS = np.random.uniform(0, 255, size=(len(CLASSES), 3))

Lines 20-23 build a list called CLASSES containing our labels. This is followed by a list, COLORS which contains corresponding random colors for bounding boxes (Line 24).

Now we need to load our model:

# load our serialized model from disk
print("[INFO] loading model...")
net = cv2.dnn.readNetFromCaffe(args["prototxt"], args["model"])

The above lines are self-explanatory, we simply print a message and load our model (Lines 27 and 28).

Next, we will load our query image and prepare our blob , which we will feed-forward through the network:

# load the input image and construct an input blob for the image
# by resizing to a fixed 300x300 pixels and then normalizing it
# (note: normalization is done via the authors of the MobileNet SSD
# implementation)
image = cv2.imread(args["image"])
(h, w) = image.shape[:2]
blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 0.007843,
	(300, 300), 127.5)

Taking note of the comment in this block, we load our image (Line 34), extract the height and width (Line 35), and calculate a 300 by 300 pixel blob from our image (Line 36).

Now we’re ready to do the heavy lifting — we’ll pass this blob through the neural network:

# pass the blob through the network and obtain the detections and
# predictions
print("[INFO] computing object detections...")
net.setInput(blob)
detections = net.forward()

On Lines 41 and 42 we set the input to the network and compute the forward pass for the input, storing the result as detections . Computing the forward pass and associated detections could take awhile depending on your model and input size, but for this example it will be relatively quick on most CPUs.

Let’s loop through our detections and determine what and where the objects are in the image:

# loop over the detections
for i in np.arange(0, detections.shape[2]):
	# extract the confidence (i.e., probability) associated with the
	# prediction
	confidence = detections[0, 0, i, 2]

	# filter out weak detections by ensuring the `confidence` is
	# greater than the minimum confidence
	if confidence > args["confidence"]:
		# extract the index of the class label from the `detections`,
		# then compute the (x, y)-coordinates of the bounding box for
		# the object
		idx = int(detections[0, 0, i, 1])
		box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
		(startX, startY, endX, endY) = box.astype("int")

		# display the prediction
		label = "{}: {:.2f}%".format(CLASSES[idx], confidence * 100)
		print("[INFO] {}".format(label))
		cv2.rectangle(image, (startX, startY), (endX, endY),
			COLORS[idx], 2)
		y = startY - 15 if startY - 15 > 15 else startY + 15
		cv2.putText(image, label, (startX, y),
			cv2.FONT_HERSHEY_SIMPLEX, 0.5, COLORS[idx], 2)

We start by looping over our detections, keeping in mind that multiple objects can be detected in a single image. We also apply a check to the confidence (i.e., probability) associated with each detection. If the confidence is high enough (i.e. above the threshold), then we’ll display the prediction in the terminal as well as draw the prediction on the image with text and a colored bounding box. Let’s break it down line-by-line:

Looping through our detections , first we extract the confidence value (Line 48).

If the confidence is above our minimum threshold (Line 52), we extract the class label index (Line 56) and compute the bounding box around the detected object (Line 57).

Then, we extract the (x, y)-coordinates of the box (Line 58) which we will will use shortly for drawing a rectangle and displaying text.

Next, we build a text label containing the CLASS name and the confidence (Line 61).

Using the label, we print it to the terminal (Line 62), followed by drawing a colored rectangle around the object using our previously extracted (x, y)-coordinates (Lines 63 and 64).

In general, we want the label to be displayed above the rectangle, but if there isn’t room, we’ll display it just below the top of the rectangle (Line 65).

Finally, we overlay the colored text onto the image using the y-value that we just calculated (Lines 66 and 67).

The only remaining step is to display the result:

# show the output image
cv2.imshow("Output", image)
cv2.waitKey(0)

We display the resulting output image to the screen until a key is pressed (Lines 70 and 71).

OpenCV and deep learning object detection results

To download the code + pre-trained network + example images, be sure to use the “Downloads” section at the bottom of this blog post.

From there, unzip the archive and execute the following command:

$ python deep_learning_object_detection.py \
	--prototxt MobileNetSSD_deploy.prototxt.txt \
	--model MobileNetSSD_deploy.caffemodel --image images/example_01.jpg 
[INFO] loading model...
[INFO] computing object detections...
[INFO] loading model...
[INFO] computing object detections...
[INFO] car: 99.78%
[INFO] car: 99.25%
Figure 3: Two Toyotas on the highway recognized with near-100% confidence using OpenCV, deep learning, and object detection.

Our first result shows cars recognized and detected with near-100% confidence.

In this example we detect an airplane using deep learning-based object detection:

$ python deep_learning_object_detection.py \
	--prototxt MobileNetSSD_deploy.prototxt.txt \
	--model MobileNetSSD_deploy.caffemodel --image images/example_02.jpg 
[INFO] loading model...
[INFO] computing object detections...
[INFO] loading model...
[INFO] computing object detections...
[INFO] aeroplane: 98.42%
Figure 4: An airplane successfully detected with high confidence via Python, OpenCV, and deep learning.

The ability for deep learning to detect and localize obscured objects is demonstrated in the following image, where we see a horse (and it’s rider) jumping a fence flanked by two potted plants:

$ python deep_learning_object_detection.py \
	--prototxt MobileNetSSD_deploy.prototxt.txt \
	--model MobileNetSSD_deploy.caffemodel --image images/example_03.jpg
[INFO] loading model...
[INFO] computing object detections...
[INFO] horse: 96.67%
[INFO] person: 92.58%
[INFO] pottedplant: 96.87%
[INFO] pottedplant: 34.42%
Figure 5: A person riding a horse and two potted plants are successfully identified despite a lot of objects in the image via deep learning-based object detection.

In this example we can see a beer bottle is detected with an impressive 100% confidence:

$ python deep_learning_object_detection.py --prototxt MobileNetSSD_deploy.prototxt.txt \
	--model MobileNetSSD_deploy.caffemodel --image images/example_04.jpg 
[INFO] loading model...
[INFO] computing object detections...
[INFO] bottle: 100.00%
Figure 6: Deep learning + OpenCV are able to correctly detect a beer bottle in an input image.

Followed by another horse image which also contains a dog, car, and person:

$ python deep_learning_object_detection.py \
	--prototxt MobileNetSSD_deploy.prototxt.txt \
	--model MobileNetSSD_deploy.caffemodel --image images/example_05.jpg 
[INFO] loading model...
[INFO] computing object detections...
[INFO] car: 99.87%
[INFO] dog: 94.88%
[INFO] horse: 99.97%
[INFO] person: 99.88%
Figure 7: Several objects in this image including a car, dog, horse, and person are all recognized.

Finally, a picture of me and Jemma, the family beagle:

$ python deep_learning_object_detection.py \
	--prototxt MobileNetSSD_deploy.prototxt.txt \
	--model MobileNetSSD_deploy.caffemodel --image images/example_06.jpg 
[INFO] loading model...
[INFO] computing object detections...
[INFO] dog: 95.88%
[INFO] person: 99.95%
Figure 8: Me and the family beagle are corrected as a “person” and a “dog” via deep learning, object detection, and OpenCV. The TV monitor is not recognized.

Unfortunately the TV monitor isn’t recognized in this image which is likely due to (1) me blocking it and (2) poor contrast around the TV. That being said, we have demonstrated excellent object detection results using OpenCV’s dnn module.

Alternative deep learning object detectors

In this post, we used OpenCV and the Single Shot Detector (SSD) model for deep learning-based object detection.

However, there are deep learning object detectors that we can apply, including:

Additionally, if you are interested in learning how to train your own custom deep learning object detectors, including obtaining a deeper understanding of the R-CNN family of object detectors, be sure to read this four-part series:

  1. Turning any CNN image classifier into an object detector with Keras, TensorFlow, and OpenCV
  2. OpenCV Selective Search for Object Detection
  3. Region proposal object detection with OpenCV, Keras, and TensorFlow
  4. R-CNN object detection with Keras, TensorFlow, and Deep Learning

From there, I recommend studying the concept of bounding box regression in more detail:

  1. Object detection: Bounding box regression with Keras, TensorFlow, and Deep Learning
  2. Multi-class object detection and bounding box regression with Keras, TensorFlow, and Deep Learning

What's next? We recommend PyImageSearch University.

Course information:
86+ total classes • 115+ hours hours of on-demand code walkthrough videos • Last updated: May 2025
★★★★★ 4.84 (128 Ratings) • 16,000+ Students Enrolled

I strongly believe that if you had the right teacher you could master computer vision and deep learning.

Do you think learning computer vision and deep learning has to be time-consuming, overwhelming, and complicated? Or has to involve complex mathematics and equations? Or requires a degree in computer science?

That’s not the case.

All you need to master computer vision and deep learning is for someone to explain things to you in simple, intuitive terms. And that’s exactly what I do. My mission is to change education and how complex Artificial Intelligence topics are taught.

If you're serious about learning computer vision, your next stop should be PyImageSearch University, the most comprehensive computer vision, deep learning, and OpenCV course online today. Here you’ll learn how to successfully and confidently apply computer vision to your work, research, and projects. Join me in computer vision mastery.

Inside PyImageSearch University you'll find:

  • ✓ 86+ courses on essential computer vision, deep learning, and OpenCV topics
  • ✓ 86 Certificates of Completion
  • ✓ 115+ hours hours of on-demand video
  • ✓ Brand new courses released regularly, ensuring you can keep up with state-of-the-art techniques
  • ✓ Pre-configured Jupyter Notebooks in Google Colab
  • ✓ Run all code examples in your web browser — works on Windows, macOS, and Linux (no dev environment configuration required!)
  • ✓ Access to centralized code repos for all 540+ tutorials on PyImageSearch
  • ✓ Easy one-click downloads for code, datasets, pre-trained models, etc.
  • ✓ Access on mobile, laptop, desktop, etc.

Click here to join PyImageSearch University

Summary

In today’s blog post we learned how to perform object detection using deep learning and OpenCV.

Specifically, we used both MobileNets + Single Shot Detectors along with OpenCV 3.3’s brand new (totally overhauled) dnn module to detect objects in images.

As a computer vision and deep learning community we owe a lot to the contributions of Aleksandr Rybnikov, the main contributor to the dnn module for making deep learning so accessible from within the OpenCV library. You can find Aleksandr’s original OpenCV example script here — I have modified it for the purposes of this blog post.

In a future blog post I’ll be demonstrating how we can modify today’s tutorial to work with real-time video streams, thus enabling us to perform deep learning-based object detection to videos. We’ll be sure to leverage efficient frame I/O to increase the FPS throughout our pipeline as well.

To be notified when future blog posts (such as the real-time object detection tutorial) are published here on PyImageSearch, simply enter your email address in the form below.

Download the Source Code and FREE 17-page Resource Guide

Enter your email address below to get a .zip of the code and a FREE 17-page Resource Guide on Computer Vision, OpenCV, and Deep Learning. Inside you'll find my hand-picked tutorials, books, courses, and libraries to help you master CV and DL!

About the Author

Hi there, I’m Adrian Rosebrock, PhD. All too often I see developers, students, and researchers wasting their time, studying the wrong things, and generally struggling to get started with Computer Vision, Deep Learning, and OpenCV. I created this website to show you what I believe is the best possible way to get your start.

Reader Interactions

Previous Article:

Raspbian Stretch: Install OpenCV 3 + Python on your Raspberry Pi

Next Article:

Real-time object detection with deep learning and OpenCV

510 responses to: Object detection with deep learning and OpenCV

  1. tommy

    how do we train the dnn using opencv or do we have to use tensorflow and the likes?

    plus where can we get some sample caffemodels?

    tensorflow has some models in its own ckpt format.

    • Adrian Rosebrock

      I would start by giving the first post in the series a read. You do not train the models with OpenCV’s dnn module. They are instead trained using tools like Caffe, TensorFlow, or PyTorch. This particular example demonstrates how to load a pre-trained Caffe network.

      The dnn module has been totally re-done in OpenCV 3.3. Many Caffe models will work with it out-of-the-box. I would suggest taking a look at the Caffe Model Zoo for more pre-trained networks.

      • Chinh

        Hi Adrian,

        I need detect object on the traffic. Do you have a new topic about Caffe models? How to create or retrain model to custom caffe model for my way?

  2. Max

    Hi Adrian,
    how long does it take to forward walk through the provided network?
    Is it faster than tensorflow based networks of same architecture?
    Is there a tutorial inside of your books that covers fast recognition and detection using CNN at best in realtime with networks like YOLO.

    • Adrian Rosebrock

      1. As I’ll be discussing in next week’s tutorial you’ll be able to get 6-8 frames per second using this method.

      2. Once the model is trained you won’t see massive speed increases as it’s (1) just the forward pass and (2) OpenCV is loading the serialized weights from disk.

      3. Yes, I will be covering object detection inside Deep Learning for Computer Vision with Python. You’ll want to go with the ImageNet Bundle where I discuss SSD and Faster R-CNNs.

  3. Vasanth

    Hi Adrian , You always inspired me with your Tremendous Innovation and become my Role Model too….

    Now Coming back to the Topic , I’m Getting this error :

    Traceback (most recent call last):
    File “deep_learning_object_detection.py”, line 32, in
    net = cv2.dnn.readNetFromCaffe(args[“prototxt”], args[“model”])
    AttributeError: ‘module’ object has no attribute ‘dnn’

    Eventhough after installing Lasagne , it is giving me the error :
    ImportError: Could not import Theano.

    Please make sure you install a recent enough version of Theano. See
    section ‘Install from PyPI’ in the installation docs for more details:
    http://lasagne.readthedocs.org/en/latest/user/installation.html#install-from-pypi

    • Adrian Rosebrock

      Hi Vasanth — you need to install OpenCV 3.3 for this tutorial to work. Lasange and Theano are not needed and you can safely skip them.

      • David Crawley

        Is there any way to make this work with OpenCV 3.2 – I am trying to make this work with ROS (Robot operating system) but this only incorporated OpenCV 3.2. AM I SOL don’t go there territory or is there a way?

        • Adrian Rosebrock

          Hey David — I wish I had better news for you. The dnn module was completely and entirely overhauled in OpenCV 3.3. Without OpenCV 3.3 you will not have the new dnn module and therefore you cannot apply object detection with deep learning and OpenCV.

          Again, I hate to be the bearer of bad news.

          • Rodrigo Passos

            I upgraded to 3.3.0:
            pip install –upgrade opencv-python
            or python -m pip install –upgrade opencv-python

          • Adrian Rosebrock

            Be careful when doing this — you’ll be missing out on additional libraries and you may not have GUI support.

      • Usama

        Hi Adrian,

        I tried installing opencv 3.3 but I am still getting the same issue below:

        PS python
        deep_learning_object_detection.py –image images/example_01.jpg –
        -prototxt MobileNetSSD_deploy.prototxt.txt –model MobileNetSSD_de
        ploy.caffemodel
        [INFO] loading model…
        Traceback (most recent call last):
        File “deep_learning_object_detection.py”, line 32, in
        net = cv2.dnn.readNetFromCaffe(args[“prototxt”], args[“model”])
        AttributeError: ‘module’ object has no attribute ‘dnn’

        Any suggestions?

  5. aditya

    Can you please provide the dataset link and the train.py file
    i want to manually train it and check it…
    So please provide the dataset name or downloading link and the program to train the model…

  6. Sydney

    Nice tutorial. Can i please have the video implementation of the object detection method. The challenge i am facing is of the model using up all my resources for inference and i am sure this method goes a long way in ensuring efficient resource usage during inference.

  7. Terry

    God send you to save my life. I struggled for months about the performance issue with yolov2. It’s just too heavy for cpu and mobile devices.

  9. Chris Albertson

    I’m still trying to understand how an image classifier cold be incorporated into a larger network for find bounding boxes. I thought about searching a tree of cropped images buy that would be interactive and slow.

    I looks like this article took the black-box approach. How to detect objects? Make a call to an object detector. That’s easy but how does the object detector work?

    How can an object classifier like vgg16 be used for deception without iteration

    • Adrian Rosebrock

      Traditional object detection is accomplished using a sliding window an image pyramid, like in Histogram of Oriented Gradients. Deep learning-based object detectors do end-to-end object detection. The actual inner workings of how SSD/Faster R-CNN work are outside the context of this post, but the gist is that you can divide an image into a grid, classify each grid, and then adjust the anchors of the grid to better fit the object. This is a huge simplification but it should help point you in the right direction.

  10. Barbara

    Hi Adrian, how can I edit your code to only detect person? The others shapes aren’t necessary for me. And thank you so much for your tutorial, it helps a lot

    • Adrian Rosebrock

      The “person” class is the 14th index in CLASSES and therefore the returned detections as well. You can remove the for loop that loops over the detections and then just check the probability associated with the person class:

      confidence = detections[0, 0, 14, 2]
if confidence > threshold:
    ...

      • Barbara

        It didn’t work. the detections return only the shapes that were detected. if I had only 2 shapes in my image, the for loop will repeat twice, then integration would be 0 and 1 and not the whole CLASSES. So, your answer is wrong. I’ve tried it. But I can’t find a way of detecting only human shape.

        • Adrian Rosebrock

          Try this:

          for i in np.arange(0, detections.shape[2]):
    confidence = detections[0, 0, i, 2]
    idx = int(detections[0, 0, i, 1])

    if idx == 14 and confidence > threshold:
        print("person")

          You’ll want to double-check that the idx is indeed 14.

          • Barbara

            That’s is exactly what I tried, but it’s 15 for “person”. You said in other comment that you’d be sharing the video implementation on Monday. I already did that following the instructions here and others about video. But, it takes around 17 s between frames (between processing a frame and another). Do you know what I could do to decrease this time?

          • Adrian Rosebrock

            Hi Barbara — unfortunately without knowing more about your setup I’m not sure what the issue is. I would kindly ask you to please wait until the video tutorial is released on Monday, September 18th. There are additional optimizations that you may not be considering such as reducing frame size, using threading to speedup the frames per second rate, etc.

          • Manuel

            I modified this algorithm to find only people, there are many false positives.
            Is it possible to integrate it with a face search? I just want to know if there is a person in the picture, not a position in the picture, recognition or something else.

          • Adrian Rosebrock

            Are you trying to detect the presence of a face in an image? Simple Haar cascades or HOG + Linear SVM detectors could easily accomplish this. Take a look at this blog post as well as Practical Python and OpenCV for help with face detection.

            If you’re trying to actually recognize the face in an image you should use face recognition algorithms such as Eigenfaces, Fisherfaces, LBPs for face recognition, or even deep learning-based techniques. The PyImageSearch Gurus course covers face recognition techniques.

            I hope that helps!

  11. siam

    after running that code i found that error:argument -i/–image is required
    How can I fix it?
    I am using windows 10, and python 2.7

  13. Jose fernando

    hello adrian I am from Colombia you would recommend using linux for a higher performance or no problem if you use windows Thanks

    • Adrian Rosebrock

      I would definitely recommend using Linux for deep learning environments. macOS is a good fallback or if you’re just playing around and learning fundamentals. I would not recommend Windows.

  17. Flávio Rodrigues

    Hi, Adrian. Have you tried the original TensorFlow Model to compare with the Caffe version? Do you plan to do such tests and show on your blog how to use a pre-trained model with differentt Network architectures? Thanks a lot for your great posts. It encourages me even more to buy your books, and I hope I will!

    • Adrian Rosebrock

      I personally haven’t benchmarked the original TensorFlow model compared to the Caffe one; however, the author of the TensorFlow did benchmark them. They share their benchmarks here and note the differences in implementation.

      I’ve already covered how to use GoogLeNet and now MobileNet in this post. I’ll cover more networks in the future. Otherwise, for a detailed review of other state-of-the-art architectures (and how to implement them) I would definitely refer you to Deep Learning for Computer Vision with Python.

      • Flávio Rodrigues

        Thanks a lot, Adrian. And I have just watched your new real-time object detection video on YouTube. Oh, man, stop blowing my mind! Hahaha. I can’t wait to see the blog post. And thank you for always answering our questions. You must be a super organized person to do that on such a busy schedule. Cheers.

  18. Alan Federman

    Traceback (most recent call last):
    File “deep_learning_object_detection.py”, line 32, in
    net = cv2.dnn.readNetFromCaffe(args[“prototxt”], args[“model”])
    AttributeError: ‘module’ object has no attribute ‘dnn’

    I missed a step somewhere.

    • Adrian Rosebrock

      Hi Alan — it looks like you do not have OpenCV 3.3 installed. Please ensure OpenCV 3.3 has been installed on your system.

  19. Gilad

    Hi Adrian,
    I tried to combine this code with your previous code which uses googlenet, but found out that the forward procedure doesn’t support localization.
    If I don’t care about the computation timing and would like to have much more classes with localization, what should I do?
    Thx,
    G

    • Adrian Rosebrock

      Unfortunately in that case you would need to train your own custom object detector to on the actual ImageNet dataset so you can localize the 1,000 specific categories rather than the 20 that this network was trained on.

  20. Scott Stoltzman

    Is there a list out there of the different “classes” that can be detected? I have searched extensively and can’t find anything. My guess is that there are A LOT of them.

    • Adrian Rosebrock

      Hi Scott — please see this blog post, specifically Lines 20-23. The CLASSES list provides the list of classes that can be detected using this pre-trained network.

    • Adrian Rosebrock

      Yes, just be sure to perform the calibration step via the triangle similarity (as discussed in the “Measuring distance between objects in an image” post you linked to).

  22. computernut

    Have you had a chance to look at the Neural network on a stick from Modivus? (developer dot movidius dot com/ ) Do you believe if it holds promise for this sort of application, where small and faster computation is more the need than the crunching power of say the Nvidiai Tesla machines?

    • Adrian Rosebrock

      It really depends on how well Intel documents the Movidius stick (Intel isn’t known for their documentation). The Movidius is really only meant for deploying networks, not training.

      • Flávio Rodrigues

        Hi, Adrian. Maybe it’s something worth to give it a try. The stick is not that expensive and appears to increase the frame rate substanttialy on a Pi 2 or 3. I’m waiting for your post about real-time object detection on a Pi, but I’m afraid that it doesn’t work so well. I have seen these two videos (https://www.youtube.com/watch?time_continue=4&v=f39NFuZAj6s ; https://www.youtube.com/watch?v=41E5hni786Y) and i’m wondering how would it be using such pre-trained Caffe models running on Movidius NCS with a Raspberry Pi and OpenCV. It would be awesome! Have you ever thought about exploring it?

        • Adrian Rosebrock

          I’ve mentioned the Movidius in a handful of comments in other blog posts. The success of the Movidius is going to depend a lot on Intel’s documentation which is not something they are known for. I’ll likely play around with it in the future, but it’s primarily used for deploying pre-trained networks rather than training them. Again, it’s something that I need to give more thought to.

  25. rmb

    Your tutorials are really excellent! You get the impression that everything is so simple.

    On the basis of your code, which works perfectly, I would now like to identify (car / van / small trucks / large trucks).

    As you suggested, I looked into the Caffe Model Zoo. I tried to use GoogLeNet_cars by retrieving directly .model (http://mmlab.ie.cuhk.edu.hk/datasets/comp_cars/googlenet_finetune_web_car_iter_10000.caffemodel)

    And the corresponding prototxt (https://gist.github.com/bogger/b90eb88e31cd745525ae#file-deploy-prototxt)

    But simply changing the model does seem to be the right way to go. What should I do? … Yes I completely discover the subject.

    Thanks in advance.

    • Adrian Rosebrock

      You can use pre-trained models to detect objects in images; however, these pre-trained models must be object detectors. The GoogLeNet model is not an object detector. It’s an image classifier. The version of GoogLeNet you supplied cannot be used for object detection (just image classification).

      I hope that helps!

  26. Gerardo

    Interestingly, running your code on my machine gives different object detection results than yours. For instance, on example 3, I can only detect the horse and one potted plant. On example 5 I get the same detection plus the dog is also detected as a cat (with a higher probability) and the model is able to capture the person in the back, left side near the fence.

    Is this variation expected? I would have expected that the dnn model would behave the same on an the same image for all repetitions of the experiment.

    thanks for the great post!

    • Adrian Rosebrock

      There will be a very tiny bit of variation depending on your version of OpenCV, optimization libraries, system dependencies, etc.; however, I would not expect results to vary as much as you are seeing. What OS and versions of libraries are you running?

      • Peter

        Hi Adrian, I got the same result as Geraro and feel confused. there is a probability for cat with higher probability but without the box for it

        … terminal output removed to formatting …

        • Adrian Rosebrock

          Hi Peter, thanks for the comment. I’m honestly not sure what the problem is here. I have not run into this issue personally and I’m not sure what the problem/solution is. I will continue to look into it.

        • Codetonium

          There will also be variations based on the confidence level you specify. For example in the image above one plant is 89% confidence and the other is much lower at only 34%.

  27. Ravi Teja

    Hi Adrian,

    Thanks for writing wonderful tutorials. What is the best place to learn about all functions inside OpenCV module and Tensorflow deep learning modules? For understanding your code, I feel i should brushup these things first, I can better understand your code.

    • Adrian Rosebrock

      Can you elaborate on what you mean by “all functions”? If you wanted to learn about “all functions” you would read through the documentation for OpenCV and TensorFlow.

      However, I don’t think this is a very good way to learn. Instead, you should go through Practical Python and OpenCV and Deep Learning for Computer Vision with Python which teaches you how to use these functions to solve actual problems.

      Reading the documentation can be helpful to clarify the parameters to a function, but it’s not a very good way to practically learn the techniques.

  29. Zig

    Adrian,

    I’m getting the following error when trying to run your code:

    [INFO] loading model….
    …
    Can’t open “MobileNetSSD_deploy.prototxt.txt” in function ReadProtoFromTextFile

    Any idea what this could be? OpenCV 3.3, Python 3.6 (same error on 2.7). Similar error is produced when I change the model or prototxt.

    Cheers!

    • Adrian Rosebrock

      Please see my reply to “zhang xue” and confirm whether you’ve used the “Downloads” sections of this post to download the pre-trained model files.

  30. Aniket

    Hi Adrian,

    I have come across some problems when understanding your code:

    In this line,

    detections.shape[2]

    what does this line means when the blob is forward pass through the network in the line “net.forward”?

    In this line,

    confidence = detections[0, 0, i, 2]

    what are these 4 parameters(0,0,i,2) means and how it extracts the confidence of the object detected?

    In this line,

    idx = int(detections[0, 0, i, 1])

    what is this 1 signifies in detections[ ]?

    In this line,

    box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])

    what do you want to do by multiplying numpy array with detections? Why you take 4th argument of detections[ ] as 3:7, what does this mean? Why you pass [w, h, w, h] to numpy array and why you pass width and height two times to numpy array?

    Please help, thanks in advance.

    • Adrian Rosebrock

      The detections object is a mulit-dimensional NumPy array. The call to detections.shape gives us the number of actual detections. We can then extract the confidence for the i-th detection via detections[0, 0, i, 2]. The slice 3:7 gives us the bounding box coordinates of the object that was detected. We need to multiply these coordinates by the image width and height as they were relatively scaled by the SSD.

      Take a look at the detections NumPy array and play around with it. If you’re new to NumPy, take the time to educate yourself on how array slices work and how vector multiplies work. This will help you learn more.

  31. Zig

    Hi Adrian,

    Just to make sure I’m understanding what is going on here. SSD is an object detector that sits on top of an image classifier (in this case MobileNet). So, technically, one can switch to a more accurate (but slower) image classifier such as Inception. And this would improve the detection results of SSD. Is this correct? I guess I can look at your other posts about using Google LeNet and change a few lines in this example to switch MobileNet with Google LeNet in OpenCV?

    Also, have you come across any implementations or blog posts that discuss playing around with various image classifiers + SSD in Keras to perform object detection?

    Thanks once again for your blog posts. They have saved me hours and hours of time and the hair on my head.

    Cheers!

    • Adrian Rosebrock

      This is a bit incorrect. In the SSD architecture, the bounding boxes and confidences for multiple categories are predicted directly within a single network. We can modify an existing network architecture to fit the SSD framework and then train it to recognize objects, but they are not hot swappable.

      For example, the base of the network could be VGG or ResNet through the final pooling layers. We then convert the FC layers to CNV layers. Additional layers are then used to perform the object detection. The loss function then minimizes over correct classifications and detections. A complete review of the SSD framework is outside the scope of this post, but I will be covering it in detail inside Deep Learning for Computer Vision with Python.

      There are one or two implementations I’ve seen of SSDs in Keras and mxnet, but from what I understand they are a bit buggy.

      • Zig

        Will the ImageNet Bundle of “Deep Learning for Computer Vision with Python” cover code (at least to some extent) to play around with object detectors and image classifiers, like I asked in my first post? There’s plenty of stuff on the net to train image classifiers but not much if one wants to couple object detection with everything. Cheers. (Oh, and when will the review of SSD and everything related be available for reading and exploring in your book?)

        • Adrian Rosebrock

          Yes, you are absolutely correct. the ImageNet Bundle of Deep Learning for Computer Vision with Python will demonstrate how to train your own custom object detectors using deep learning. From there I’ll also demonstrate how to create a custom image processing pipeline that will enable you to take an input image and obtain the output predictions + detections using your classifier.

          Secondly, I will be reviewing SSD inside the ImageNet Bundle. I won’t be demonstrating how to implement it, but I will be discussing how it works and demonstrating how to use it.

  32. Justice

    Hi, I was wondering if I would be able to only detect fruits and vegetables and differentiate the different types?

    • Adrian Rosebrock

      Using the pre-trained network, no. You can only detect objects that the network was already trained to recognize.

      If you want to recognize custom objects (such as fruits and vegetables) you’ll need to either (1) train a new network from scratch or (2) apply transfer learning, such as fine-tuning.

  33. Justice

    Would you be able to send any helpful tools or links that would help me start the train the network from scratch?.

  34. zhang xue

    Traceback (most recent call last):
    File “deep_learning_with_opencv.py”, line 34, in
    net = cv2.dnn.readNetFromCaffe(args[“prototxt”], args[“model”])
    cv2.error: /home/ubuntu/opencv-3.3.0/modules/dnn/src/caffe/caffe_io.cpp:1113: error: (-2) FAILED: fs.is_open(). Can’t open “MobileNetSSD_deploy.prototxt.txt” in function ReadProtoFromTextFile

    how to solve it?thanks

    • Adrian Rosebrock

      Just to clarify, have you used the “Downloads” section of this blog post to download the source code + pre-trained Caffe model and prototxt files?

        • Adrian Rosebrock

          Hi Jason — thanks for the comment. I’ve seen a handful of readers run into this problem. Unfortunately I have not been able to replicate it. It would be a big help to me and the rest of the PyImageSearch community could help to replicate this error.

          • Fahad

            Hi Adrian, thanks for all of your efforts in making such useful tutorials. I am also facing the same error though I have downloaded the code from the website.

            “Can’t open “MobileNetSSD_deploy.prototxt” in function cv::dnn::ReadProtoFromTextFile”

          • Fahad

            I guess I have found the solution, at least it worked for me. Some times downloaded files are blocked by the computer, so you have to open the properties of Model file and Prototxt file, and check the UNBLOCK at bottom right. Hopefully it would work. Thanks again.

          • Hung Tran

            Hi Adrian. Thanks a lot for your tutorials.
            But i’m still having this error. Have you found any ways to solve it ?

          • Adrian Rosebrock

            Hi Hung Tran — what operating system are you using? I would suggest double-checking the paths to your prototxt and model files.

          • Alue

            Hi, Thank you for this nice tutorial.I have another question.
            Traceback (most recent call last):
            File “deep_learning_object_detection.py”, line 43, in
            net = cv2.dnn.readNetFromCaffe(args[“prototxt”], args[“model”])
            cv2.error: /Users/travis/build/skvark/opencv-python/opencv/modules/dnn/src/caffe/caffe_io.cpp:1122: error: (-2) FAILED: fs.is_open(). Can’t open “MobileNetSSD_deploy.caffemode” in function ReadProtoFromBinaryFile
            How to solve this?Thank you!

          • Adrian Rosebrock

            Please double-check the path to the input model path controlled by the --model switch. The path you supplied does not exist. That is why OpenCV is throwing an error.

          • Lillie

            Hi Adrian,

            I thought it might help to add my thought to this thread, because I had the same error, but my problem might have been very different.

            I’m also on Windows 10, using Idle in Python 3.8.

            After checking the processes using print statements, I realized that I was getting stuck in the While loop and repeatedly getting dimensions, then passing to blob.

            There reason being that everything after ‘detections = net.forward()’ was outside out the While loop.

            Hopefully this helps if someone encounters it again.

    • arezoo

      Hi! thank u for this nice tutorial. I have the same error and I download the code from this site. Fahad solution does not work for me… I’m using python 3.5 , opencv 3.3 on windows 8..
      Any suggestions?

    • arezoo

      I Found the solution for this problem and it worked for me…
      the problem was in addressing the args … i entered the whole path in CMD and it worked…
      i did s.th like this :

      python d:\object-detection-deep-learning\deep_learning_object_detection.py -i d:\object-detection-deep-learning\images\example_01.jpg -p d:\object-detection-deep-learning\MobileNetSSD_deploy.prototxt.txt -m d:\object-detection-deep-learning\MobileNetSSD_deploy.caffemodel -c 0.2

      hope it works for you!

    • Aditya Rai

      Just add .txt after the protxt filename while giving the parameters if you are using Windows.
      ie: –protxt C:\Users\rai_a\PycharmProjects\sdd\MobileNetSSD_deploy.prototxt.txt

  35. pavi111

    what algorithm you used detect object in image or can you please links for research paper , other code for object detection from image in which i can train my own images as it will be covered in your book you told but for now i need a reference as a part of my project….
    so i would be glad if you can share github link for thr object detection code with train.py file.

    Thanks your tutorials are too good….

  36. Aniket

    Hello Adrian,

    I want to play with this code on my pc which is windows 7 64-bit. On my machine, I still don’t yet have opencv installed and even I don’t know about which configuration(working environment) should I have in order to run this code. I even don’t know how to install opencv on my pc so that this code will run, please help…..

    • Adrian Rosebrock

      Hi Aniket — if you are interested in studying computer vision and deep learning I would recommend that you use either Linux or macOS. Windows is not recommended for deep learning or computer vision. I demonstrate how to configure Ubuntu for deep learning and macOS for deep learning.

      Otherwise, I offer a pre-configured Ubuntu VirtualBox virtual machine as part of my book, Deep Learning for Computer Vision with Python.

      This VM will run on Windows, macOS, and Linux and is by far the fastest way to get up and running with deep learning and OpenCV.

      I hope that helps!

  38. JohnZ

    Hi Adrian,
    first of all, thanks for this great tutorial!

    I have a short question: I am trying to rebuilt your tutorial with the openCV C++ API. When I see the call for the function for the blog generation from the input image:

    cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 0.007843, (300, 300), 127.5)

    it is hard for me to match it up with the corresponding C++ API function

    Could you give me a small hint how to match it? Escpecially the scalar value “0.007843” and “127.5” did not realy match for me.

    Thanks for you help and again great work!
    Johannes

    • Adrian Rosebrock

      I’ll actually be doing a tutorial that details every parameter of the cv2.dnn.blobFromImage in the next few weeks. In the meantime, 127.5 is the mean subtraction value and 0.007843 is your normalization factor.

      • JohnZ

        Hi Adrian, thanks for your fast reply.

        Ok, is this a special function you are using? I am currently using openCV 3.3 from august this year. Actually, I do not understand yet how the normalization factor fits to the current API. There is the mean value which gets subtracted from each color channel and parameters for the target size of the image. And finally a boolean flag to swap the red and green channels.

        Could you give me a hint please?

        • Adrian Rosebrock

          You are correct. The mean value is computed across the training set and then subtracted from each channel of the image. You can also optionally supply a 3-tuple if you have different RGB values (which in most cases you do). Once you perform the mean subtraction you multiply by the scaling value.

          • JohnZ

            Ok, thanks for the hint. Sorry for bothering you again but would be this call correct?

            cv::Mat inputBlob = cv::blobFromImage(img, 0.007843, cv::Size(300, 300), cv::Scalar(127.5));

            Again, thank you for your great tutorials!
            Johannes

          • Adrian Rosebrock

            I have only used the Python bindings of the “dnn” module, not the C++ ones. It looks like your call is correct, but again, you should compile your code and try it.

        • Durgesh Trivedi

          Hi John,

          I am trying to convert the python code in C++ , I think you already did it . is it possible for you to share it or give some direction on it.
          I am trying to detect just 1 object . I am able to run the c++ example provided by OPenCV but want to add the rectangle around the object . I am not so good in python so unable to understand much out of it.

  39. Nihit

    Hello,
    I was trying to replicate your results of example 3. In my case only the horse and potted plants were getting detected and not the person. Either I had to remove the mean (127.5) from blobFromImage or resize to 400×400 to get person detected. Do you know why so ?

    • Adrian Rosebrock

      Hi Nihit — that is indeed strange; however, I’m not sure why that would be. Did you use the “Downloads” section of the post to use the same code, pre-trained network, and example images that I used?

        • Adrian Rosebrock

          Thank you for sharing the additional details, Nihit! Unfortunately I’m not sure what the exact issue is here. I wish I could help more, but without physical access to your machine to diagnose any library issues, I’m not sure what the problem may be.

        • Big Sky

          Did you solve this? Same issue here. Tried opencv versions 3.3.0.9, 3.3.0.10, 3.3.1.11, 3.4.0.12, 3.4.0.14, 3.4.1.15, 3.4.2.16, 3.4.2.17. All have this issue.

  40. Jes

    Hi! Thanks for the clear tutorial, really makes difference in trying to figure this stuff out!
    This is what I don’t get about how the dnn works (I’m a newbie with the object detection so :D):
    how does the model go through the blob to get the location? I mean, if the object recognition model is (presumably) trained with the object nicely framed in the middle of the image, how does the detection model find a small or partially covered object like the baseball glove? Does it somehow divide the image in seqments?

    • Adrian Rosebrock

      The model is not trained with images that have the objects nicely framed in the center of the image. Instead, images are provided with plaintext bounding boxes that indicate where in the image the object is. The SSD then learns patterns in the input images that correspond to the class labels while simultaneously adjusting the predicted bounding boxes.

      If you’re new to computer vision and object detection be sure to read this post on the fundamentals on more traditional object detectors.

  41. Mustafa Demir

    Hi, thanks for this post but I have a problem.
    error: AttributeError: module ‘cv2.cv2’ has no attribute ‘dnn’

    • Adrian Rosebrock

      So this is either (1) a typo or (2) you haven’t installed OpenCV 3.3.

      The correct call is cv2.dnn., not cv2.cv2.dnn.

      Secondly, please ensure you have installed OpenCV 3.3 on your system.

  42. Emy

    Hi, thanks for this post but I have a problem.
    after running that code i found that error:argument -i/–image is required
    How can I fix it?

  43. Paul Kuo

    Hi, Adrian,

    say that I have a GPU card fitted in my machine, would opencv dnn module utilizes it to speed up the detection and how would it do it? Thanks ~~

    • Adrian Rosebrock

      As far as I understand, Python cannot access the GPU-bindings for OpenCV. I would suggest taking a look at the C++ API of OpenCV.

      • Paul Kuo

        Cool, thank you for your suggestion. As my projects are all developed with C++ openCV APIs, this will be easier for me if the opencv C++ APIs could access the GPU-bindings.

        Also I am looking forward to your next post regarding object detection on a video stream~~

        thanks

        Paul

  44. Prabhat Kumar Prabhakar

    Traceback (most recent call last):
    File “real_time_object_detection.py”, line 33, in
    net = cv2.dnn.readNetFromCaffe(args[“prototxt”], args[“model”])
    AttributeError: ‘module’ object has no attribute ‘dnn’

    this is the issue i am getting while running the real_time_object_detection.py file
    is there anything wrong with my opencv installation??

    I installed it from the link provided by you, i din run into any issue , while installation, however while running the real_time_object_detection.py file i get the above error.

    Please help, anyone if came across such issue

    • Adrian Rosebrock

      Please read the comments before posting. I’ve already addressed this issue multiple times. Take a look at my reply to “Vasanth”. Please ensure you have properly installed OpenCV 3.3.

  45. Satyam

    Hey Adrian,

    Thank you so much for making such great tutorials.Just wanted to know the way to train the model for a huge database (for more objects other than listed in the classes).

    Thank you!

  46. Ahmad

    Hi Adrian,

    usage: deep_learning_object_detection.py [-h] -i IMAGE -p PROTOTXT -m MODEL
    [-c CONFIDENCE]
    deep_learning_object_detection.py: error: argument -i/–image is required

  47. Dixon Dick

    Adrian,

    I just downloaded the source and the images were supposed to be in the .ZIP but I don’t see them. Not sure if I fumbled, but downloaded a second time and still not in the .ZIP. Not sure what I am doing wrong.

    dcd

  49. Rahul

    Hi Adrian,
    Thanks for the great tutorial.
    I have used it for object detection, and it works like charm on my laptop!!.

    I tried to replicate the same thing on Jetson Tx1, on which Opencv4Tegra was preinstalled, but, while installing Opencv 3.3 make -j4, there arises space issues.
    It seems i do not have enough space on Tx1.

    Can you suggest what are the possible options that might get me out of this problem,
    Thanks in Advance.

    • Adrian Rosebrock

      I would suggest using an external SD card, or better yet, external drive.

      Be sure to download the OpenCV repos to your SD card/external drive and do the compile there. This will ensure you have the additional space during the compile. After the compile has finished run sudo make install which will copy the compiled files to their appropriate locations.

  50. Sebastian

    Hi,

    I really enjoyed your tutorial because it gave me a good start with this interesting topic. So one question regarding object detection. Is there an approach that will tell me, if a general object is in my image or not? Let’s say we have a background that stays the same and there is an object in the image. I do not know what is the object, but that an object is there. At the moment I tend to solve this problem with “classic” computer vision, is there a deep learning approach? Maybe check if no object is matched? (with certain probability)

    • Adrian Rosebrock

      The .prototxt files are similar to configuration files for Caffe. They are in plaintext format and specify either (1) how to train a model or (2) the architecture of the model.

  53. Tahlil Ahmed Chowdhury

    Hi Adrian. Thanks very much for this awesome tutorial. I have one concern here though. You only take one image for detection. But this is not efficient. If I have multiple images or a video file I could read bunch of images/frames and try to detect them all at once. That will be much faster. This is a huge problem that I’m facing right now with RCNN. I can test one image but I could not find any solution how to do batch testing. It would be really great if you could also do a post about it.

    Love your work btw 🙂 Thanks very much.

  54. Pooja chavan

    Amazing post !..really inspired me to work on my computer vision project . I am a new baby and was really worried about it. Thank you !..I would try this and move for my project !..we were going to fine tune vgg16 with google ref dataset…which machine learning library do you suggest for use ?

    • Adrian Rosebrock

      Are you trying to perform object detection or image classification? Keep in mind that VGG16 cannot be directly used for object detection. You would need to fit into a deep learning + object detection framework, such as SSD.

  55. Ton ten Kate

    Nice example. I looked at the script of Aleksandr Rybnikov you mentioned in the post and tried to adapt your example to use it with the tensorflow prelearned model.
    I adapted it to the 90 classes, used the ssd_mobilenet_v1_coco.pbtxt from opencv_extra and downloaded ssd_mobilenet_v1_coco_11_06_2017.tar.gz to get the frozen_inference_graph.pb.
    At first I used the graph.pbtxt included in the tar file but that doesn’t work with OpenCV3.1.1 and your script. So I tried the ssd_mobilenet_v1_coco.pbtxt from opencv_extra. This sort of works (doesn’t give errors) but the object recognition results are not good.
    Is there a way to generate an OpenCV3.1.1 compatible *.pbtxt? to work with your script or doesn’t it work this way?

  56. wally

    I built OpenCV 3.3 on a Raspberry Pi3 following your Raspbian Stretch instructions and downloaded this sample code. Everything seems to work except my results don’t quite match the results shown in this blog.

    Particularly example_05.jpg where I get:
    [INFO] car: 99.49%
    [INFO] cat: 61.79%
    [INFO] dog: 50.56%
    [INFO] horse: 99.80%
    [INFO] person: 86.79%
    [INFO] person: 26.94%

    Instead of what you show:
    [INFO] car: 99.87%
    [INFO] dog: 94.88%
    [INFO] horse: 99.97%
    [INFO] person: 99.88%

    Seems I should get the same results with the same code and test images, but it appears I don’t. The “boxes” drawn on my images seem better located than those in your example, except for the cat, which is not really there and probably is drawn over by the box for the dog.

    I setup virtual environments for python3 and python2.7 and my results are the same with the python3 and python2.7 environment, but different from yours.

    • Adrian Rosebrock

      Hey Wally — sorry for any confusion here, but I updated the code in the blog post to provide better localization. That is the reason why your results do not 100% match up with mine.

  57. Rabbani

    Hi Adrian!
    I was trying this with an input image containing a bat and a ball. Since these classes aren’t part of the trained classes, I was expecting that the classifier doesn’t classify my image into any classes. However it was classifying the bat and the ball into ‘Aeroplane’ and ‘Bottle’.
    Is there any way through which the classifier doesn’t classify an image containing untrained objects and instead outputs a message saying that the classifier was not able to detect any classes.

    • Adrian Rosebrock

      There are “background” classes (i.e., “not interesting objects” or “not an object all”) that are used when training some object detectors; however, these only work in some contexts. I would suggest upping the minimum probability used to filter out weak predictions.

      • Rabbani

        Hi! I was not able to detect the ‘background’ class, even when testing it against ‘white background’ image! Could you provide me with some idea for an image where the background class can be detected.
        Secondly I wanted to ask if the training file is available for this? I wanted to train some classes on my own.
        Thirdly is there any portal where datasets for multiple images can be easily availible that can be used to test this?
        I am hoping to receive your guidance at the earliest. Thank you so much 🙂

        • Adrian Rosebrock

          1. I haven’t played with the background class for this model.

          2. Please see my reply to “Justice” on September 27, 2017.

          3. Are you referring to the “background” class?

  58. Shrunoti Karpe

    Any chance I can label these detected objects, so I can distinguish between two same type of objects. For example, if I have two dogs and distinguish them as dog1 and dog2

    • Adrian Rosebrock

      Can you go into a bit more detail? Are you monitoring a video stream and you want to be able to track multiple objects (in this case, dogs)?

  59. Ziga

    Dear Adrian!
    Thank you for a kind example. I am new to neural networks and I am wondering, how much speedup can one achieve, if the object classification is trained only for e.g. aeroplane in comparison to this case, where the detector is trained for 20 classes? Is that 2 times, 3 times, certainly not 20 times? What is your rough estimate?

    What about the size of the training file – this should be reduced 20-times?

    Thank you for your effort and for educating us!

    • Adrian Rosebrock

      The number of classes a network has to recognize does not change the size of weights in the network (within reason). What changes the size of the network and associated weight file is the depth and number of parameters. You can use the same architecture and use 20 classes or 2 classes and the output model would be almost identical and size. Again, it’s the depth and type of architecture. I discuss this more in my book, Deep Learning for Computer Vision with Python.

  60. Noor khokhar

    usage: work.py [-h] -p PROTOTXT -m MODEL [-c CONFIDENCE]
    work.py: error: the following arguments are required: -p/–prototxt, -m/–model
    >>>

    what should i do? please guide me.
    i am using python 3.5.2 and opencv 3.3

    • Adrian Rosebrock

      It sounds like you are trying to parse your command line arguments inside a shell. Instead, execute the code directly via the command line as i do in the post:

      $ python deep_learning_object_detection.py \
	--prototxt MobileNetSSD_deploy.prototxt.txt \
	--model MobileNetSSD_deploy.caffemodel --image images/example_02.jpg

      Notice how I am executing the command from my command line, not with the Python shell itself.

      Secondly, make sure you read up on command line arguments before continuing.

  61. Damla

    Hello,
    Thank you for providing us these useful and important things.
    I have a question about dataset which we are using while training. I want to create a dataset which consists of luggage: handbag, backpack, suitcase etc.
    Does it matter to have different types of an object in the dataset, because I want to combine all these types in luggage class? Will it effect my accuracy?

    • Adrian Rosebrock

      I would run a test using different classes and all of them combined. Handbags, backpacks, and suitcases can vary quite dramatically but without seeing your particular dataset my gut tells me that you should be using separate classes.

  62. Sarma

    Hi Adrian,

    Thanks again for the great post.

    I am using the above code, to get distance value from rectified stereo left and right images. I detect same object in both left and right images using the cv2.dnn.blobFromImage. Then from difference in the horizontal pixel location,i am finding distance.

    But the blob returns different vertical pixel values for same object, as the images are rectified, we should get same value right, do you know why this happens ?

    Also the estimated distance is erroneous, is is due to resize or scaling that we apply during cv2.dnn.blobFromImage function ?

    Thanks in advance !

  63. young smith

    Dear Adrian!
    Thank you for a kind example. I am finding a way of using a video as a input. how can i do this? please, help me:)

    • Adrian Rosebrock

      Please take a look at this blog post where I discuss performing object detection in real-time using deep learning. Instead of supplying the index of the webcam to cv2.VideoCapture you can pass in a file path. If you’re new to using OpenCV for video processing I would suggest reading through my introductory book, Practical Python and OpenCV. I hope that helps!

  64. Jacqueline Garcia

    I don’t know if this question will be answered or if anyone will know how to answer. But I am getting video feed from a TurtleBot Kinect Camera. How would I go about showing the feed with the rectangles if I am using the TurtleBot’s Kinect camera???

  69. Amare Mahtsentu

    thanks for real for your unlimited support of the community
    I need to use SSD person detection In a transport vehicle in their sitting position.
    do I need to train the person detection or the pretrained data in Caffe is enough?
    Is there any method of counting the bounding boxes after person detection?
    does caffe limited of youtube tutorials?

    • Adrian Rosebrock

      When building a production-level system you should always train or fine-tune on images that represent what the CNN will be used to detect in real-world scenarios. I would suggest fine-tuning on your own dataset if at all possible.

    • Adrian Rosebrock

      Hi Jasper, be sure to check your OpenCV version:

      $ python
>>> import cv2
>>> cv2.__version__

      The “dnn” module is only in OpenCV 3.3 and above. My guess is that you have an earlier version of OpenCV installed and that you will need to reinstall OpenCV.

  71. Ajeya B Jois

    i downloaded some images from google and its not working ,its coming like this
    (h, w) = image.shape[:2]
    AttributeError: ‘NoneType’ object has no attribute ‘shape’

    but if i try for images given along with source code it works anybody help

  72. akshra

    Hello. what are the numbers in blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 0.007843,
    (300, 300), 127.5)?
    how did you get the 0.0078 and the 127.5?

  73. hfad

    is there any way to measure the height of the bounding boxes drawn using, say, the x and y position that the program returns?

    • Adrian Rosebrock

      The height in terms of pixels? Or the height in terms of a real-world metric such as inches, millimeters, etc.?

  74. Margi

    When I run this code then it gives error:

    usage: deep_learning_object_detection.py [-h] -i IMAGE -p PROTOTXT -m MODEL
    [-c CONFIDENCE]
    deep_learning_object_detection.py: error: the following arguments are required: -i/–image, -p/–prototxt, -m/–model

    • Adrian Rosebrock

      Please read my reply to “Ahmad” on October 28, 2017. You can search this page for “command line arguments” as well.

  75. Aayush Dua

    Hi Adrian,
    Thanks for your help in learning Mobilenet and SSD using dnn module. This blog is first of its kind and very unique. However i wanted to know if we can extend the number of classes more than 20, say 100? If so can you please guide as to how can do that?

    • Adrian Rosebrock

      You would need to:

      1. Gather example images of the additional objects you want to recognize (including any images the network was originally trained on if you wanted to continue to utilize those classes).
      2. And then either re-train the network from scratch or fine-tune it

      Deep Learning for Computer Vision with Python covers both of these methods.

  76. Rakshith

    Hi Adrian,
    Thank You! This was a beautiful guide. I have one question though, what are the 4 values in detections[] ? in line 42. You use it in the loop like this detections[0, 0, i, 2] what is 0,0,i,2?

  77. Damien

    Hey Adrian,
    I am trying to get the code from object detection and deep learning to work:
    I’ve downloaded it and when i open a jupyter notebook in the directory of the files and run the code:

    $ python deep_learning_object_detection.py \
    –prototxt MobileNetSSD_deploy.prototxt.txt \
    –model MobileNetSSD_deploy.caffemodel –image images/example_06.jpg

    I get an invalid syntax error. I’ve even tried loading the numpy, cv2 and still nothing. How to i get the code to run?
    Thank you

    • Adrian Rosebrock

      Are you trying to run the Python script from within the Jupyter Notebook? If so, that is your error. Execute the script from your command line.

  78. Joost

    What an excellent blog. My pics are size 640×480, and I see much more accurate results (detecting objects as opposed to not detecting anything at all sometimes) when I modify the source code to not resize to 300×300, (lines 36-37), but to put 640×480 there. Is this to be expected, and why yes why no? Ofcourse I should invest the time to learn exactly what it is I’m doing, but my time for these things unfortunately is limited ;(

    • Adrian Rosebrock

      So keep in mind if your images at not resized to 300×300 pixels than OpenCV will just take the center crop of your 640×480 image and then process it. Perhaps the center of your image contains higher resolution objects that you are trying to detect and using the center crop helps enable this?

      • Joost

        Yes I’m trying to detect quite small objects, relative to the image size. Why will it center crop? It cannot work with arbitrary image dimensions? What are the restrictions?

        And if it needs to be resized to 1:1 ratio (like 300×300), why is the changed aspect ratio then not an issue? (If resizing 640×480 to 300×300, for example, the aspect ratio ofcourse changes)

      • Vicent

        Hey! I have had some problems regarding the line:

        blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 0.007843,
        (300, 300), 127.5)

        That’s, I think, because “ciop” argument is not false by default, but true, so I ha problems with the detection until I added this crop=False argument.

  79. Maurad

    Hi Adrian,

    Great post as always!
    I am currently preparing a dataset to train SSD on it in order to localize my own objects.
    What is the best way to prepare data for the training and validation part:
    – is it to make annotations (class_id + bounding box) for each object in the images I have
    – or crop my images to isolate my objects alone in smaller images, and then put them in a folder which represents its class?

    Would one of these techniques make a difference during training?
    I am asking this question because I noticed that for classifiers the second method is used while for detectors the first one is used.
    But I couldn’t found anywhere if annotations were a rule for detectors or just a convention.

    As for the test images, I perfectly understand the use of annotations.
    Thanks in advance for any support you could provide me

    • Adrian Rosebrock

      You should always make annotations of the class ID + bounding boxes of each object in an image and save the annotations to a separate file (I recommend a simple CSV or JSON file). You can always use this information to later crop out bounding boxes and save the ROIs individually if you wish. The reverse is not true.

      Since SSDs and Faster R-CNNs have a concept of hard-negatives (where they take a non-annotated ROI region and see if the network incorrectly classifies it) you’ll want to supply the entire image to the network, not just a small crop of the ROI.

  80. Vaibhav Chaturvedi

    Adrian,
    I just wanted to say that I am deeply impressed by your diligence and sincerity in your blog posts.
    I have immense respect for you.
    I am an applied-maths guy and was looking to catch on recent developments in cv and your posts arrived just at the right time.

    Seriously, Thank You.

    In case I can be of any help to you ever, please do let me know.

    • Adrian Rosebrock

      Thank you Vaibhav, I really appreciate that. Comments like these really put a smile on my face and make my day 🙂

  82. Falgun

    Hi Adrian,

    Awesome blog for image detection using openCV. Thank you for this one.

    If I want to train my own set of images (not the COCO dataset) on MobileNetSSD, how can I do that ?

    My goal is to detect an object in an image, crop that object and then run a color detection on that cropped image. It would be really helpful if you could provide some help for the same 🙂

    Thank you once again.

  84. Sanjaya Kumar Das

    What algorithm has been used for only object detection in an image and what is its computational complexity?

  85. Vijay

    Thank you for the post. I am just wondering why you used a Caffe model instead of a TensorFlow model, could you please elaborate on this? Thanks!

    • Adrian Rosebrock

      OpenCV’s “dnn” module works a bit better with Caffe models right now. I’m sure in future releases of OpenCV the TensorFlow model loading will become more robust, but for the time being OpenCV supports loading Caffe models a bit better.

  86. Rituparna Das

    What is the algorithm used here for object detection and classification and what is the time and space complexity for the same?

    • Adrian Rosebrock

      The blog post discusses the algorithm used for detection: MobileNet + Single Shot Detector (SSD).

      • Eric

        Hello Adrian,

        The class labels (21 labels) used for initialization at the beginning of the code in this post are those used during the training. That’s the reason why you choose only 21 labels in the post. Am i right ?

        There are more than 21 objects in the COCO dataset. Why do we only choose 21 of them as labels ? I mean we can set, say, 100 labels during the training, of course that would require more training time.

        • Adrian Rosebrock

          The 21 labels in this post are the 21 class labels the network was initially trained on. The creator of this model trained on a subset of the full dataset.

    • Adrian Rosebrock

      Hey Ali — sorry, I haven’t used a Windows system in a good many years. I’m not sure about the best way to install Caffe on Windows.

  90. Rarat

    Hello Adrian, may u give me solution?

    i want add new object in your project, for the step , what i doing in first time ?

    thnks adrian

  91. Matt Kleinsmith

    Great intro. I didn’t read the code part because I was looking for reasoning regarding training new classes (classes outside of PASCAL VOC, or whichever dataset the pretrained weights were trained on). I look forward to reading more of your articles.

  92. Ganesh

    Hi Adrian thanks for the intro! I ran into this problem

    net = cv2.dnn.readNetFromCaffe(args[“prototxt”], args[“model”])
    cv2.error: D:\Build\OpenCV\opencv-3.4.0\modules\dnn\src\caffe\caffe_io.cpp:1119: error: (-2) FAILED: fs.is_open(). Can’t open “MobileNetSSD_deplot.prototxt” in function cv::dnn::ReadProtoFromTextFile

    Opencv is working fine in my other applications. Thank you!

    • Adrian Rosebrock

      The path to your input prototxt file is incorrect. Make sure you use the “Downloads” section of this post to download the source code and then double-check your paths to the input .prototxt and .caffemodel files.

  93. arielbendoy

    i got an error, please help!
    usage: object.py [-h] -i IMAGE -p PROTOTXT -m MODEL [-c CONFIDENCE]
    object.py: error: the following arguments are required: -i/–image, -p/–prototxt, -m/–model

    • Adrian Rosebrock

      Hey there, I’ve addressed this question a few times in the comments section. See my reply to “Noor khokhar” on December 4, 2017 to help you get started.

  94. daysi

    Hi Adrian, I have been following your posts, great stuff. I am saving money to buy one of your bundles. By the way, have you look into Keras RetinaNet implementation?, I would like to hear your thoughts.

  96. Aqsa

    Hi

    Really a very interesting Post. How can I train this software for detecting only one category like road signs from a given image .

  97. Vishwas Chandran

    error: the following arguments are required: -i/–image, -p/–prototxt, -m/–model
    An exception has occurred, use %tb to see the full traceback.

    im getting the error

    • Adrian Rosebrock

      You need to supply the command line arguments to the script, exactly as I do in the post. If you’re new to command line arguments, that’s okay, but you should read up on them first.

    • Adrian Rosebrock

      It loops over the total number of detections from the network. I would suggest adding some “print” statements in the code to help you debug and visualize this as well.

    • Adrian Rosebrock

      It performs a “forward pass” of the network. Simply put: it computes the detections and associated probabilities.

  100. Mamta

    Hi Adrian,
    I want the SSD mobilenet to classify trains, truck and other vehicle types too. Please tell me how do I add more classes/categories to the pretrained model ?

    Thanks
    M

    • Adrian Rosebrock

      You cannot directly add more classes to the pre-trained model. You would need to either train the model from scratch or apply transfer learning via fine-tuning. I discuss how to train your own custom deep learning object detectors, including how to recognize different types of vehicles, inside Deep Learning for Computer Vision with Python.

  101. pranith

    Hi Adrian,
    Your blogs have helped me understand the code easily and I thank you for that.
    What if I want to reduce the number of classes for detection?
    I have tried doing that and have been facing errors with idx out of range.
    Here are the changes I’ve made:
    CLASSES = [“bicycle”,”bus”, “car”, “motorbike”, “person”]
    And executing this I get the following error:
    Traceback (most recent call last):
    File “real_time_object_detection.py”, line 67, in
    label = “{}: {:.2f}%”.format(CLASSES[idx],confidence * 100)
    IndexError: list index out of range

    • Adrian Rosebrock

      You don’t want to modify the CLASSES list. Instead, when looping over the detected objects, use an if statement to filter out classes you are not interested in.

  102. David Mata

    Hi Adrian.

    Great post as always.

    I have one question

    What is the difference between training a convolutional neural network for classification and one for object detection?

    I know that when you train a CNN for classification you need a big dataset of images where those images contain the objects that we want the network learns to recognize, but for object detection how do you train the CNN (For example with SSD, I know it would be different if we train a YOLO network)
    The paper for SSD says “ground truth information needs to be assigned to specific outputs in
    the fixed set of detector outputs” (What it means with ground truth information needs to be assigned?)
    “Once
    this assignment is determined, the loss function and back propagation are applied endto-end.” (This is the normal training for a CNN)

    “Training also involves choosing the set of default boxes and scales for detection
    as well as the hard negative mining and data augmentation strategies.”
    (How do we apply this?)

    For me, an object detection is one which can detect an object, no matter what that object is, but it seems that a CNN for object detection can only recognize objects for what it was trained. (For example, if we train an SSD to detect objects of dogs we train the model with a dataset of dogs)
    If that is the case, I don´t see why to have 2 CNN to detect objects (1 for classification and another one for object detection)

    For what I understood in your post is that once you are ready, you have 2 models 1 for object classification y another for object classification.
    How do you combine both models to work together?

    • Adrian Rosebrock

      Simply put:

      1. A classification network will give you a class label of what the image contains.
      2. An object detection network will give you multiple class labels AND bounding boxes that indicate where in the image each object is.

      Keep in mind that it’s impossible for a machine learning model to recognize classes or objects it was not trained it. It has to be trained on the classes to recognize them.

      If you’re interested in learning more about classification, object detection, and deep learning, I would suggest taking a look at Deep Learning for Computer Vision with Python where I discuss the techniques in detail (and with source code to help solidify the concepts).

      • David Mata

        Thanks Adrian.

        So, what you are saying is that for object detection there is only one neural network that will bring the class label and the bounding boxes? I just need one big dataset and with it I can train my neural network for object detection?

        or an object detection network is form with 2 differente networks, one for class label and other for bounding boxes?

        • Adrian Rosebrock

          You’re understanding is very close but I want to clarify one point:

          You normally start with what we call a “base network”. This network is typically, but not always, pre-trained on an existing dataset for classification. We then modify the network architecture, remove some layers, add new special ones, and transform it into an object detection network. We then train the entire modified network end to end to perform detection.

  103. deekshith

    hello,
    i have downloaded openCV 3.3
    and also the code that was mailed to me.
    the problem is i dont know how to run it.
    i am new to this and i have no clue on how to go about the execution of this code and if i require any other software.
    so it would be really helpful if someone gave me steps to execute it.
    please….

  104. Tai

    First of all, love your work. And especially love this tutorial for making ML easily understandable and used with opencv.

    Just wanted to let you know about the MobileNet-SSD object detection model trained in TensorFlow found by following the information in opencv > dnn > samples > “mobilenet_ssd_accuracy.py” has alot higher accuracy (or more detections if accuracy isnt the right word here).
    It detected the tv in the background of your last picture and detected relatively small people in a picture that the caffe model provided here didnt. With roughly the same time for prediction

    https://github.com/opencv/opencv/blob/master/samples/dnn/mobilenet_ssd_accuracy.py

  105. Ankita Vaidya

    Your blogs have helped me understand the code easily and I thank you for that.If I want to detect fruits on tree specifically fruits like apple,mango,strawberry,watermelon,orange,pineapple then what should I use.
    Actually I have detected on tree fruits on the basis of color. But that is not much accurate .Is their any way to detect and identiy on tree fruit.

    • Adrian Rosebrock

      It is certainly possible to detect various fruits in an image/video; however, you will need to train your own custom object detector. I would suggest taking a look at Deep Learning for Computer Vision with Python where I provide detailed instructions (including code) on how to train your own object detectors. After going through the book I am confident that you will be able to train your fruit detector 🙂

  106. Bhavitha Maile

    Hey can you explain me the different parameters used in the layers in prototxt file and how the image is processed from one layer to other like what is the input and output of the hidden layers?
    how do we decide the number of layers?

    Also how does the entrire process goes?
    please help me out.
    thnak you.

    • Adrian Rosebrock

      Hey Bhavitha — explaining the entire process of how an image/volume is transformed layer-by-layer by a network is far too detailed to cover in a blog post comment, especially when you consider the different types of layers (convolution, activation, batch normalization, pooling, etc.).

      The gist is that a network is inputted to a network. A total of K convolutions are applied resulting in a MxNxK volume. We then pass through a non-linear activation (ReLU) and optionally a batch normalization (sometimes the order of activation and BN are swapped). Max pooling could be used to reduce volume size or convolutions can be used as well if their strides are large enough.

      This process repeats, reducing the size of the volume and increasing the depth as it passes through the network.

      Eventually we use a fully-connected layer(s) to obtain the final predictions.

      If you’re interested in learning more about CNNs, including:

      – How they work
      – The parameters used for each layer
      – How to piece together the building blocks to build your own CNN architectures

      Then I suggest you work through Deep Learning for Computer Vision with Python where I discuss all if this in detail.

      I hope that helps!

      • Bhavitha Maile

        Thanks for the suggestion Adrew.
        Is there any big difference between the CNN and MobileNets+SSd?
        what do you mean by depthwise separable convolution in detail?

        • Adrian Rosebrock

          A CNN is used for image classification. A CNN is also used as a base network in the SSD framework. When saying “MobileNet + SSD” we’re saying that MobileNet is the base network and SSD is the object detection framework.

  107. Md Alauddin

    usage: deep_learning_object_detection.py [-h] -i IMAGE -p PROTOTXT -m MODEL
    [-c CONFIDENCE]
    deep_learning_object_detection.py: error: the following arguments are required: -i/–image, -p/–prototxt, -m/–model

    this error occured

  108. Nzo

    Hi Adrian,

    Thank you for your posts. I have learned alot from yours.

    So is it right if I say we can use MobileNet base-network with YOLO framework?

  109. Abhisek

    Hi Adrain,
    Thanks a lot for this wonderful tutorial, I was trying to detect human hands from the caffe model obtained from “http://vision.soic.indiana.edu/projects/lending-a-hand/”, in case you gets time please tell me how to fix it. I tried it both your tutorial as well as opencv dnn samples (C++)

    • Adrian Rosebrock

      Hey Abhisek — I don’t have any experience with this model so I’m not sure what the error is. It does look cool though so if I have some spare time I might take a look (no promises though).

      • Nihel

        Hello,

        In your program (deep learning object detection.py) you have detected 20 object, but for me I choose to detected other electronic object as resistance, diode, Microcontroller …. I would like to help you to show how add these object.

        Thank you

        • Adrian Rosebrock

          Got it, thank you for the clarification — I understsand the question now.

          You have two options:

          1. Fine-tune an existing object detection model
          2. Train your own object detector from scratch

          You cannot simply modify the code to detect your microcontroller components — you need to train the network.

          I cover how to train your own custom deep learning-based object detectors inside Deep Learning for Computer Vision with Python. I would suggest starting there.

          I hope that helps point you in the right direction!

  111. Prasanna Kumar Routray

    Hello Adrian,
    how can I add ‘ball’ to the classes? so that I can also detect ball in the provided image.

    Thanks,
    Prasanna

    • Adrian Rosebrock

      You would need to:

      1. Gather example images of balls (see this post)
      2. Train or fine-tune an object detector on your new dataset

      You cannot simply add “ball” to the classes. It requires training or fine-tuning the network.

  112. Kartik

    Hi Adrian,

    I am running object detection on Rpi 3 with a raspicam(Raspberry pi camera connect via CSI cable) I am getting following error . I tried debugging None type error but no luck can you please help into this?

    Following is error:

    ploy.caffemodel
    [INFO] loading model…
    [INFO] starting video stream…
    Traceback (most recent call last):
    …
    (h, w) = image.shape[:2]
    AttributeError: ‘NoneType’ object has no attribute ‘shape’

    • Kartik

      PS: camera is working fine. I tried normal capture and streaming to a local web server but its working properly.

      • Adrian Rosebrock

        To start, take a look at my reply to “Ajeya B Jois December 29, 2017”. My reply discusses “NoneType” errors and how to resolve them. Additionally, the post you commented on does not include real-time object detection — perhaps you meant this post?

  113. Nilesh

    I got the detection using readNetFromDarknet() using python, but I am not able to figure out how to iterate over the detection and draw bounding box on image. Pl suggest

    I am using tiny-yolo-voc.cfg and tiny-yolo-voc.weights

    • Adrian Rosebrock

      Hi Nilesh — I have not tried using the readFromDarnket function yet. Once I do I will write a post on it.

  114. Usup Suparma

    I want to add a new class but it does not work. the example of the class I want to add is the ladder. what’s wrong with this. does the class of stairs not exist? from the reading even shows that the picture is a chair.

    • Adrian Rosebrock

      Are you using the pre-trained network in this blog post? Keep in mind that the network was never trained on a “ladder” or “chair” class. You would need to either train the network from scratch or apply fine-tuning. This appears to be a common misconception with this post so I’ll make sure to write a follow up tutorial in early May.

      • Usup Suparma

        yes. i used the pre-trained network in this blog post. but not work for ladder or chair. what do I do if I want to add new data? so I do not add data to previously trained data?

        • Adrian Rosebrock

          Hi Usup — stay tuned for my blog post that will go live on 5/14/2018.

  115. Kaustubh B

    Hello,

    Thanks for the tutorial.

    What’s basically “.caffemodel”
    and how can I create my own ?

    How do I read a .caffemodel file ?

    • Adrian Rosebrock

      Hi Kaustubh, training a Caffe model is outside the scope of this blog post. I do cover Caffe in PyImageSearch Gurus if you’re interested in learning to train Caffe models.

  116. Nomad

    Thanks for interesting material!
    But how we can to train model to recognize some new object?
    May be your have short tutorial for this?

    • Adrian Rosebrock

      Yes, this model can be fine-tuned. Do you have any experience fine-tuning models before?

  118. Amare

    Hi Adrian !!
    do you have any idea how to run SSD detector fast; like how to increase FPS? I implement SSD( single shot multi box detector) person detector and add a dlib tracker with it. but it is very very slow to the extent that it can not be used for real time applications. but Hog detectors work well with dlib tracker and it is fast (sufficient for real time apps).
    Thank you in advance

  119. Vosco

    Thanks for the inspiration Adrian. Well, I have almost followed all your blog post and successfully applied some of it. Particularly for this blog, I have investigated several related papers started from R-CNN and all the way to YOLO.
    But I have been stacking with very first stage of all of those methods, how to prepare own custom dataset include annotation so I can use my own custom dataset . It seems all those methods are using public dataset which already annotated.

    Thank you.

  120. ghizlane

    hello adrian
    how can i change the classe

    i want detect shapes of object , like circle or squarre but real things , can u please tell me how change in class

  121. Chakri

    Sir,

    Is this method follows Non-Max-Suppression? When I’m running this algorithm it is showing a person with different boxes and upper half and lower half but not the person as whole object?

    • Adrian Rosebrock

      This method does indeed apply NMS internally. In your particular image it sounds like the network is localizing the person as two objects. This could be due to an odd angle of the person in the image, the input resolution, or image quality.

  122. vishakraj

    Hello

    Thanks for the post it is really usefull
    how to get other objects to detect by using this program..
    and how to train the objects to help program to detect in the video frame..
    could you please tell me how to do it

    thanks in advance

  123. Rebzenok

    Hello. I would like to know how can I get a background from some image, and, example, to compare him with another image background?

    • Adrian Rosebrock

      I’m not sure what you mean by “compare background” in this context. Could you please elaborate?

      • Rebzenok

        For example, you have 2 pictures of one area, but a couple of people or a car are present in the first picture. And on the other picture there is, for example, a bicycle or a cat. The background remains the same (large buildings, trees), but only some moving objects have changed. And I want to compare background of this images to understand same or different this pictures

  124. Riya208

    Hey,

    First of all, I love your blog. They are simple and easy to follow.

    Second of all, I have a question. From the description above, I understand that

    –prototxt : The path to the Caffe prototxt file.
    –model : The path to the pre-trained model.

    I have installed Caffe successfully. I have OpenCV version 3.4.1 and I am using python 3.5

    So my question is:

    Does MobileNetSSD_deploy.prototxt.txt gets install when one installs Caffe? I could not find it in the “Caffe” (the installed) folder.

    Also, how do I train the model?

    For example, I want to train an image with a different set of objects (not the one mentioned above) and would like to have lesser neural network layers (since I do not have a complicated image to train). How do I do that?

    I am new to deep learning and trying to understand the program.

    Thank you very much!

    Best Regards

    • Adrian Rosebrock

      1. You actually don’t need Caffe for this example, just OpenCV 3.3+. OpenCV will load the Caffe files.

      2. No, you need to actually download the prototxt and model using the “Downloads” section of the blog post.

      3. If you wanted to train/fine-tune the model you would need to use the Caffe framework itself. If you’re interested in doing this I would recommend take a look at the PyImageSearch Gurus course as well as Deep Learning for Computer Vision with Python.

      • Riya208

        Hi,

        Thanks for the reply 🙂

        A quick question:

        How did you build the prototxt file and trained the model?

        Regards

        • Adrian Rosebrock

          You’ll want to take a look at the Caffe library. As I mentioned in my previous comment, I discuss how to train networks using Caffe inside the PyImageSearch Gurus course. That said, you might want to take a look at Keras along with the TensorFlow Object Detection API to train your own custom object detectors as well.

  126. wally

    I’ve had such good results with your realtime Movidius examples and MobileNetSSD as a starting point for adding AI “person detection” to video security systems, that I decided to use this as a starting point for a stand-alone Raspberry Pi3 & PiCamera module AI enhanced video security system.

    Basically I just replaced image = cv2.imread(args[“image”]) with image = vs.read() in the loop after starting the camera with vs = VideoStream(usePiCamera=True, resolution=DISPLAY_DIMS, framerate=8).start()

    It works wonderfully and for monitoring intrusions and the one frame every ~2 second is very useful in many situations.

    Minor problem is the latency to detection. I can walk through the FOV, sit down at my monitor and then watch the frames where I walk through be processed. I’ve reduced the framerate from the default to 8 and it made no noticeable difference. Is there an option to use VideoStream from your imutils without the threading? Or should I just switch to using picamera module directly? I’m using your latest imutils v-0.4.6.

    But the big problem is after it runs for a while it seems the net.forward() stops detecting the new image and returns the detection from the previous detection, sometimes just for the next frame a two, others continuously until I move in front of the camera again or move it to apparently force another detection. I’ve tried setting detections=0 before the net.forward() call and it makes no difference. Something weird seems to be going on in the dnn opencv module. I have two nice example images to illustrate the problem if you tell me where to upload them. (I could put then on OneDrive and post a link, if that is allowed here)

    Basically you can see a person walking briskly enter the right side of the frame and get a good detection and then the next image ~2 seconds later about to exit the left side of the frame but the detection box is drawn from the previous detection on the right side!

    My Pi3 has a heat-sink attached and the CPU/GPU temps report ~70/71 C

    • Adrian Rosebrock

      If you don’t want to use threading you should just use the picamera module directly. As for your second question, I answered that in your comment in your other comment. In the future please keep all comments related to the project on the same post. It gets too confusing to jump back and forth — and furthermore, other readers cannot learn from your comments either.

  127. vipul

    deep_learning_object_detection.py: error: the following arguments are required: -i/–image images/example_01.jpg, -p/–prototxt MobileNetSSD_deploy.prototxt, -m/–model MobileNetSSD_deploy.caffemodel

    how to solve this error??

  128. Rodrigo

    Hi Adrian, i need your help please !

    How can I edit your code to only detect cars?
    Thank you so much for your tutorial !

  129. hashir roshan

    hi adrian ,

    i got result with larger sized ouput imgae so that i could’t saw the ouput image fully. so how can i reduce the ouput image size to fit my screen

    • Adrian Rosebrock

      You should resize the image before calling cv2.imshow. You could use cv2.resize or my imutils.resize function.

  130. Artur

    Hello, Adrian! Great lesson, thank you!

    I have a question. In your post you mentioned that it example based on combination of the MobileNet architecture and the Single Shot Detector (SSD) framework. As I understood you right, it example only suit to COCO dataset and was pretrained on it.

    What if I want to use it network for my purposes? I need to gather my own database and train it network on it? If yes, what requirements to images will be, where to find it? And how to train it? Use Caffe, right?

    Just want to clarify it details and get any possible links.

    Thank you for answer.

    • Adrian Rosebrock

      You are correct that you would need to gather your own dataset but you don’t have to use Caffe (other deep learning frameworks can be used as well). I actually cover how to train your own custom object detectors inside Deep Learning for Computer Vision with Python. You should also read this guide on the fundamentals of deep learning object detection.

  131. Shivaprasad P

    I executed the code but it is not recognizing person from image example_03.jpg.

  132. Shweta

    if i have a picture with multiple persons and dogs in it, how do i come to know what is the percentage of women and the percentage of dogs in the picture ?? the percentage obtained from this technique is the accuracy rate which is not what i’m looking for .
    plz help!

    • Adrian Rosebrock

      Hey Shweta — this is a simple math problem.

      Total objects = the # of objects detected in an image with probability greater than the minimum confidence
      Percentage of women = # of women detected / total objects
      Percent of dogs = # of dogs detected / total objects

      I hope that helps!

  133. Shivam Sahil

    Hmm… I get this error all the time:
    deep_learning_object_detection.py: error: the following arguments are required: -i/–image, -p/–prototxt, -m/–model

    Even after putting the path in help= “”
    Please help and let me know where am I going wrong.
    I am running it in IDLE as python file, first downloaded then added the path and then F5 to run in shell.

    • Shivam Sahil

      Also I tried using the same way you used to command prompt but it didn’t seem to work. Am I supposed to make any changes in program? My current directory is the same where models and images are stored. and btw I am using windows cmd

    • Adrian Rosebrock

      The post you are referring to uses Haar cascades which can be very hard to tune the parameters to on an image-by-image basis. You should consider using a deep learning detector (like the one covered in this post).

  135. SKR

    Hey Adrian, thanks for this wonderful article and for so many comments; I went through each of them and spawned multiple tabs. I have three questions as below.
    1) Are you aware of any trained dataset which consist of primitive geometrical shapes viz. squares, circles, rectangles, semi-circles, quadrilaterals, polygons, etc. where the shapes are just wire-frames and not the solid types filled with some colors?

    2) If such a dataset exists then can deep learning like in this article be applied to recognize multiple shapes of different sizes stacked together in a drawing as in https://imgur.com/a/5yw1b2m ? If yes, can there sizes be extracted too using some technique? For some reason I don’t want to use openCV image processing functions but want to apply deep learning in this problem.

    3) If such a dataset DOES NOT exist then is the following strategy feasible/practical?
    a) Collect all the primitive and non-primitive shapes of different sizes occurring in many such drawings and put them into a dataset and annotate them manually.
    b) Train a model through some popular technique on this dataset to learn the features.
    c) Use this model with deep learning combined with openCV and detect shapes.

    I deeply appreciate you taking out time to read through this and for providing any advices, pointers, or existing work. Many thanks.

    • Adrian Rosebrock

      Of the top of my head, sorry, I do not know of such a dataset. But you could easily create one yourself using OpenCV’s built-in drawing functions. Loop over random selections of shapes, sizes, colors, etc. and then create your dataset that way. From there you can train your own model.

      • SKR

        Thanks a lot Adrian, it saved my time in exploring such a dataset. Actually, there exist many 2D shapes dataset but they are very big and contain many different things so probably ill-suited for my problem. So what I understand from your advice is that I generate all the shapes with different parameters and create a dataset of all shapes occurring in my drawings. After that I can train a model for this dataset and do object recognition using deep learning ? Is deep learning the only solution if I want to have an AI based solution? Thanks a lot.

        • Adrian Rosebrock

          Your understanding is correct. Deep learning is certainly not the only solution but you will need to leverage machine learning to some extent. I think you would be a great fit for the PyImageSearch Gurus course where I discuss image classification in detail. I have no doubt you would be able to solve your project after working through the course.

  136. aim

    hi,
    thanks a lot for this amazing tutorial. its really very helpful. i am able to execute this code on windows and getting good results. but when the same code i am executing on Raspberry Pi, i get this following error:

    (h, w) = image.shape[:2]
    AttributeError: ‘NoneType’ object has no attribute ‘shape’

    the location for this error it shows as in …/imutils/sitepackage/convenience.py line no 69

    please let me know what should i exactly change in the code. i have referred the blog written for this(https://pyimagesearch.com/2016/12/26/opencv-resolving-nonetype-errors/)
    but still i am getting the same error.

    I am looking forward for your help/inputs.

    Thanks a lot..!!! 🙂

    • Adrian Rosebrock

      Double-check the path to your input image. The error is likely due to the “cv2.imread” function returning “None” due to a non-existent file path.

  137. Katherine

    Thanks a lot for your tutorial, it worked perfectly for me.
    I have a couple of questions though.
    1) I tried the tensorflow framework itself to implement the same task – object detection, I used their example https://github.com/tensorflow/models/tree/master/research/object_detection
    It worked well but much, much slower and, what was more important to me, it consumed a tremendous amount of memory (around 1G for just one image processing). How would you explain that OpenCV does its work much better (faster, less greedy)?
    2) Do I understand properly that I can feed cv2.dnn any other supported model from other frameworks like tensorflow?

    Thanks a lot!

    • Adrian Rosebrock

      Which model did you use from the TensorFlow Object Detection API? Keep in mind that the architecture you used will have a very different memory footprint. Secondly, the cv2.dnn module does support a number of different frameworks but you’ll need to check the documentation depending on which specific one you want to use.

  138. Rupak

    Hi Adrian,

    I need detect object on the Online Shopping Page. The images could be Baby’s items, Apparel, Electronics, Home Decor etc. Do you suggest a suitable Caffe models?

  139. Vikran

    Hi Adrian,

    I am new to deep learning am i’m trying some program for face detection using opencv. Is there any pre-trained module from which we can find out the parts of body?

  140. khaled

    Hi thanks for this interresting topic, i wonder if it is possible to have the x, and y of the frame surlighting the object to know its position in the photo
    thanks again

  142. Siladittya Manna

    Modified your code to capture video from the webcam on my laptop, and run detection on each frame. Works really cool!!
    Your tutorials really are awesome!!

    • Adrian Rosebrock

      Thanks Siladittya! 🙂 And congrats on running the object detection script on your laptop.

  143. Shital

    Hey can someone help me with this.
    Is there any step by step tutorial for running this code.
    I’m totally new to this.
    Please help me out.

    • Adrian Rosebrock

      If you’re new to Python and programming in general, that’s okay, but this is a more advanced guide and it does assume you know the fundamentals. I would suggest investing some time into learning the basics of Python before trying to run these more advanced examples.

  144. Thejus Palooran

    Hi Adrian ,
    I’m Getting this error :

    Traceback (most recent call last):
    File “deep_learning_object_detection.py”, line 32, in
    net = cv2.dnn.readNetFromCaffe(args[“prototxt”], args[“model”])
    AttributeError: ‘module’ object has no attribute ‘dnn’

    Iam using opencv-3.4.2

  145. Thejus Palooran

    I’ve gone through all comments here.
    Its showing AttributeError: ‘module’ object has no attribute ‘dnn’
    I am using opencv 3.4.2 in raspberry pi 3.

    • Thejus Palooran

      >>> import cv2
      >>> cv2.__version__
      ‘2.4.9.1’

      This is what I am getting when I check for the version. Does this mean that the opencv version in 2.4.9.1? Is that different from opencv version or both are the same? Because the folder shows “opencv3.4.2”

        • Thejus Palooran

          Hello Adrian, I used your tutorials for installation.
          However the problem got fixed when I used workon cv before executing this code.
          Thank You.

        • Thejus Palooran

          Although I am getting this error

          [INFO] loading model…
          [INFO] computing object detections…
          [INFO] car: 99.95%
          [INFO] car: 95.62%
          Traceback (most recent call last):
          File “deep_learning_object_detection.py”, line 74, in
          cv2.imshow(‘Output’, image)
          cv2.error: OpenCV(3.4.2) /home/pi/opencv-3.4.2/modules/highgui/src/window.cpp:632: error: (-2:Unspecified error) The function is not implemented. Rebuild the library with Windows, GTK+ 2.x or Carbon support. If you are on Ubuntu or Debian, install libgtk2.0-dev and pkg-config, then re-run cmake or configure script in function ‘cvShowImage’

          • Adrian Rosebrock

            You would need to train your own custom model for x-ray images. Furthermore, semantic segmentation may be a better option as well.

          • Thejus Palooran

            Hi Adrian,

            Where can I get more information on training my own custom model for x-ray images? Is it possible if I buy your book?

          • Adrian Rosebrock

            I don’t have any specific tutorials on x-ray images and semantic segmentation. I am considering doing more semantic segmentation posts and perhaps even additional chapters in the future though! Be sure to signup for the PyImageSearch Newsletter to be notified when any new chapters or posts are published 🙂

  146. Vishnu A.V

    Hi, I want to detect some particular objects only which are not in this trained model from webcam, how can I do that ? please

  147. Melvin Ng

    Hi Adrian, thank you for the tutorials. I am a beginner and your tutorials are of great help. I do have some questions.

    1) I tried dnn-mobilenet ssd, using the 20 classes trained by chuanqi (same as yours). It works. But there are any pre-trained models that I can use? I am actually trying to detect boxes but sadly the 20 classes did not include boxes.

    2) Inception V3’s model do have cartons. But are they strictly for image classification? Can we use the model for object detection? If can, how can we do that?

    • Adrian Rosebrock

      1. What kinds of boxes? 2D representations that are squares? Or actual packing boxes?

      2. You would want to apply transfer learning via fine-tuning with the Inception V3 model as your “base” or “backbone” network. I discuss how to apply transfer learning for object detection inside Deep Learning for Computer Vision with Python.

    • Adrian Rosebrock

      OpenCV provides Java/Android bindings but you will need to refer to the documentation for them. I do not have experience with the Java/Android OpenCV bindings.

  150. iqbal

    deep_learning_object_detection.py: error: the following arguments are required: -i/–image, -p/–prototxt, -m/–mode
    how can i fix it
    plz help??

  153. Piotr

    Thank you very much for a great post!

    I downloaded the code and it works really well. However, I obtain slightly different detection results than the ones you showed. For instance, one potted plant and the person are missing in my detections in the file example_03.jpg (a horse jumping over a hurdle). I also get different bounding boxes in the first image of two cars on the highway.

    My question is: was the model retrained in the meantime? What I find surprising is that it seems significantly less accurate than the YOLO network you presented recently.

    Many thanks in advance for your answer!

  154. Yadnyesh

    Hey Adrain great post
    I have a question though what does the percentage(%) denote after detecting an object?

  155. Eric

    I thought COCO had 90 classes. I see this model has only 20, what happened to the other classes? Thanks Adrian.

    • Adrian Rosebrock

      There are various subsets of the COCO dataset, both existing and ones that can be manually created. This one was only trained on 20, at least according to the person who trained the model.

  156. Nikita

    Hey Adrian, great tutorials, even for beginers!
    How can I change the program to detect only people (or any single class)?
    And is it possible to crop the area, where detected object is? Im going to create traffic lights detector and find out the signal (red light or green), so i need to crop the area, where traffic light was found.
    Thank you in advance for your answer!

  157. Ahmed Khalaf

    Like usual, great job Adrian! This is just amazing! On my HP machine, when I use YOLO it takes approximately somewhere between 1 to 10 minutes to detect objects in the dog.jpg image, while running this code takes anywhere between 0.5 to 1 second for the same exact image!!! Its really amazing!

  158. Ahmed Khalaf

    Adrian,

    Kindly, I would love to figure out a way to have this program to process detection on multiple images inside Images folder automatically, and keeps running so that whenever I throw an image inside the Images folder it will process it. Then saves all outputs in a seperate output foldr!
    How can achieve that? Do you have any tutorials or some resources that guide me to accomplish this goal?

    Thanks

    • Adrian Rosebrock

      There are a few ways to accomplish that, but it’s honestly not really a computer vision question, it’s more of a Python programming question. There are plenty of tools you can use to monitor a directory for new files. Python’s watchdog seems like a good option but I Haven’t personally used it.

  159. Yadnyesh

    Hey Adrian
    Great post
    I have to only detect humans in a video what all changes do I have to make to the classes in order to detect on humans?

  160. shiko

    Hello,

    Should we change arguments like –image? sry I’m a beginner. Idk what to do to make it run.

  161. Irina

    Hi!
    I know that if we use Keras ResNet50 in kaggle kernel, we shoud load it directly from internet. Is there such a possibility to do with OpenCV models?

    • Adrian Rosebrock

      Sorry, I’m not sure what you are asking. Are you asking if Keras models are compatible with OpenCV?

      • Irina

        If we use the OpenCV model (for example MobileNet SSD), we should download it from GitHub to our computer, and then load it into our program from the directory where we have downloaded it. I meant that maybe we could skip downloading it to our computer and load it into our program directly from GitHub. This seems like a stupid question now, I’m sorry.

  163. Krishnan

    I want to scan for images of houses that have something like “Ring” door bell or a placard in front of houses, purely for tracking purposes. Model datasets defined in this article was pre-defined. How I can expand my dataset or totally ignore the dataset and have something of my own (especially for purposes I stated earlier)

  165. Roohi

    Hi,

    Thanks for this great tutorial.

    Kindly try to help me out also …

    I want to detect hexagon shape from image(s) and image(s) consists of many objects !!!

    Can you please guide me !!!

    Thanks in advance !!!

  166. Walid

    Thanks a lot.
    if I have a Caffe prototxt file that uses GPU for inference, it does not work

    Am i missing something?

    • Adrian Rosebrock

      I’ve mentioned this in the comments section a few times (so please make sure you’re reading them) but OpenCV’s “dnn” module does not yet support many GPUs, even if you compile OpenCV with GPU support. If you need GPU support for inference right this second you should be using the “pycaffe” bindings of Caffe.

  167. Priyanka

    I am reading your blogs since last few days and my interest in ML / DL has increased.

    And i have become fond of your blogs !

    Just one question about this article.

    Is CAFFE model a must ? Can we eliminate CAFFE model and opt for TensorFlow only?

  168. Aparna

    Hi Adrian,
    Your tutorials are amazing and very helpful for even a beginner like me. I’ve got 2 questions

    #1
    I am running this program on windows (i5 processor) using a 24 MP USB webcam and the speed of obtaining subsequent frames and object detection is significantly slow.Is it because of the processor and camera?Is there any way to increase the speed ?

    #2
    How to reduce misclassification?

    Thanks in advance.

  169. Stevie

    Hi Adrian, I used your object detection and set a counter so whenever it goes to a person, it’ll say person detected: 1 however, it keeps adding so it goes 1, 2, 3, 4, 5 even though there’s one person. Is there anyway I can only count one person based off the frame?

  170. Gaurav

    hey adrian i have 2 problems
    1)when i use the another picture which were not given by you then only specific part of the image is goes under the detection remaining part is not displayed on the frame output why????

    2) can you detect objects by using webcam? how?

  171. Shaheen

    If i wanted to detect say only person from the above given code what would i need to modify in the code

  172. Venkatesh

    Hello Rosebrock,

    I wanna detect a car logo and define (Car brand name) as label.
    Landmark detection algorithm. Can you help me with that?

  173. Karroy

    Hello, I want to know if this model is trained by yourself according to the steps of MobileNet-SSD, or you directly use the contents of the GitHub link, without any modifications.This answer is very important for me to understand the article and the article in the next video stream. If the model used here is your own training model, can you elaborate on your training steps?I am very eager to look forward to your reply.

  174. Mice toe

    hi Adrian,

    actually i need the code which counts the cars passed from camera ..
    like a person counting system in mall so can you guide me please.

    • Adrian Rosebrock

      You can build exactly that using this post and then swapping out the person class for the vehicle-related classes the SSD was trained on.

  176. krishnama senapathi

    How to train this model for custom dataset like i want this model to detect the objects in the sea.Can you please help me with this??

  177. Krati Dave

    Hello sir
    I want to know how to code for detecting litter on beach in python and can i connect python code with arduino ??

  178. Noor

    Hello Adrian,thank you very much for the effort that you do and especially for sharing with us this interesting things 🙂
    I want to ask you if i want to detect others object what i must change in this code ? ( i’m new in this domain)
    In my case i want to detect boxes (in stock inventory) and i tested many codes but it doesn’t work for me do you have any suggestion please ? Thank you very much

  179. Michael

    Hey Adrian
    Thank you for explaining things easily. One question, how many hidden layers used for the network?

  180. Don

    Hey Adrian,

    I would like to display the time taken for the model to detect the objects for image input. What are the necessary changes I need to make to the code?

    • Adrian Rosebrock

      Pseudocode would look something like this:

      start = time.time()
output = net.forward(blob)
end = time.time()
diff = end - start

      I hope that helps!

  182. ana

    Hello Adrian,

    When you passed the blob in neural network, is the neural network mobilenet or Mobilenet SSD itself?

    • Adrian Rosebrock

      The model in this tutorial? It’s a Single Shot Detector (SSD) with a MobileNet backbone.

      • ana

        So,i read your blog on blob image function and i’m trying to relate it here. So,the image was actually passed in a CNN first .Then, created a blob out of it….and then the blob is passed through the Mobilenet SSD neural network right?

        • Adrian Rosebrock

          1. First we load an image from disk
          2. We then create a blob from the image
          3. The blog is passed through MobileNet + SSD
          4. We obtain our bounding boxes

  184. AP

    Hi Adrian,
    I wanted to ask how can we evaluate other parameters for our models used, such as required processing power, runtime analysis, etc.? These are necessary for translating these models to production. Also, Could you suggest other metrics or factors which might be necessary for the production side of these models on a large scale? It would be great if you could share some resources for the same as well.

    • Adrian Rosebrock

      Thanks for the suggestion. I don’t have any tutorials on that but I may consider it in the future, especially for the upcoming embedded computer vision book.

  185. A Lalith Kumar

    Hi Adrian,
    Your tutorials are nice but I didn’t understand how to write .prototxt and .caffemodel files. I have searched a lot on internet but i couldn’t find any thing related to.

  186. nitish

    Sir, can i train custom objects like a tanker,chair,bags etc ,using this code ? please guide me on this.

  187. Muhamad shodri

    Hello Mr Andrian Thank you for sharing this article. I would like to ask about something about object detection.

    I have case that want to classify the person holds snack or not. If the person holds snack will give the label “person bring snack” and if not give the label “person not bring snack”. And the question is. how do i train data? i will train data separately. for instance in this case there are two objects : person and snack and i give the label which is person and which is snack.

    Or
    I create the the training data label “person holds image” with no separately?

    Thank you in advance

  188. Dylan jimenez

    Hello friend, you think it is possible to recognize defective bottles of the type with holes or deformed, obviously that these defects are notoriously visible. either with yolo or ssd?

  189. Arjun

    Is it possible to detect an object which is not present in the pre trained model classes?
    For example, I am trying to detect a few fuses in a circuit. If it is possible, could you give a brief insight on it?

  190. Edgar

    This guide is very helpful thanks a lot! How can I improve the object detection in night/dark conditions? Should I need to train a new model ?

    • Adrian Rosebrock

      Yes, but make sure you have example images that are night/dark/low contrast — your model cannot recognize objects in environments it was not trained on so make sure your training data is representative of your test data.

  191. Saurabh

    Hello Adrian,

    Thanks for writing and sharing interesting blogs.

    I am looking for “How to train SSD based object detection on the custom dataset?”. Could you please provide a pointer?

    Thanking you!

  192. CY CHEW

    After the object is detected, the class and percentage are labeled in the frame. The percentage shown in the frame is IoU?

  193. JC

    Hi Adrian,

    Thanks for the wonderful post. I learned a lot from it.
    However, can i double check with you whether I have understood the “frameworks” correctly?

    For my understanding, the MobileNet, VGG, GoogleLeNet etc. are all some “base framework” and caffe, YOLO, SSDs etc. are so call “object detection framework”. And we are connecting these two frameworks together to get the whole network to achieve the object detection task.

    If I understood the above concept correctly, then why don’t we just make one big framework to do the object detection task instead of combining two together?

    Thanks

    • Adrian Rosebrock

      Your understanding is fundamentally correct. We train “backbone networks” or “base networks” on large image datasets first. They are then used inside object detection networks.

  194. Bill Son

    Hello Adrian, can we actually find the actual object coordinates like the middle point of the object?

    • Adrian Rosebrock

      You mean the center (x, y)-coordinates? If so, yes. Take the width and height of the bounding box, divide them by two, then add in the respective starting (x, y)-coordinates of the bounding box. That will give you the center coordinates.

  196. Vijay Thammanna

    Hi,

    Thanks for the helpful tutorials. Currently i am working on detecting person sitting in a room using Raspberry Pi 3, i have used IGNORE to ignore the remaining classes other than Person. Even after ignoring “idx = int(detections[0, 0, i, 1])” is returning index value of chair, table etc, is there a way to get only person idx.
    Also when person is looking towards camera or standing are detected but if a person is not facing towards camera or sitting on the chair are not detected. Need your support to complete this assignment.

    • Adrian Rosebrock

      When you loop over the class labels you should check to see if the index exists in your IGNORE set. If so, discard the detection and continue processing the remaining detections.

  197. Ayesha Sirkhail

    Hi Adrian,
    Good job done. I am very new to this platform and a beginner. I am working on a project to detect surface imperfections of metal objects e.g. barrels/pipes etc. I am wondering if i can use this method presented by you for object detection to detect defects like pitting/ cracks etc? pls comment if its possible and how. your help is greatly needed. regards!

  198. Heartlin Sijitha

    Dear Sir,
    Thank you for such a informative blog, indeed you triggered my eagerness towards deep learning, computer vision, opencv etc
    thank you for that sir.
    Sir, I have a doubt, Will the technique mentioned in this blog detect the human toy models in showroom as “human”??
    Thank you in advance!!!!

  199. Beerus

    Thanks Dr Adrian for all your tutorials. But i have a question why the sound in the video disappear after applying the code ???

    Thanks you !!!

  201. Ayush Panchratan

    How to replace Tensorflow model in place of caffe model and make it run on it with SSD ?

  202. Uwe Rosebrock

    Hi Adrian, I have read your well written tutorials with interest. However, I was very surprised to see your surname – that we share. As there are not that many of us – I am curious.
    You can delete this comment after reading if you wish as it is irrelevant to the context- but i appreciate an email. Cheers
    Uwe

Trackbacks

Comment section

Hey, Adrian Rosebrock here, author and creator of PyImageSearch. While I love hearing from readers, a couple years ago I made the tough decision to no longer offer 1:1 help over blog post comments.

At the time I was receiving 200+ emails per day and another 100+ blog post comments. I simply did not have the time to moderate and respond to them all, and the sheer volume of requests was taking a toll on me.

Instead, my goal is to do the most good for the computer vision, deep learning, and OpenCV community at large by focusing my time on authoring high-quality blog posts, tutorials, and books/courses.

If you need help learning computer vision and deep learning, I suggest you refer to my full catalog of books and courses — they have helped tens of thousands of developers, students, and researchers just like yourself learn Computer Vision, Deep Learning, and OpenCV.

Click here to browse my full catalog.

Similar articles

You can learn Computer Vision, Deep Learning, and OpenCV.

Get your FREE 17 page Computer Vision, OpenCV, and Deep Learning Resource Guide PDF. Inside you’ll find our hand-picked tutorials, books, courses, and libraries to help you master CV and DL.

Already a member of PyImageSearch University? Click here to login.

Access the code to this tutorial and all other 500+ tutorials on PyImageSearch

Enter your email address below to learn more about PyImageSearch University (including how you can download the source code to this post):

What's included in PyImageSearch University?

  • Easy access to the code, datasets, and pre-trained models for all 500+ tutorials on the PyImageSearch blog
  • High-quality, well documented source code with line-by-line explanations (ensuring you know exactly what the code is doing)
  • Jupyter Notebooks that are pre-configured to run in Google Colab with a single click
  • Run all code examples in your web browser — no dev environment configuration required!
  • Support for all major operating systems (Windows, macOS, Linux, and Raspbian)
  • Full access to PyImageSearch University courses
  • Detailed video tutorials for every lesson
  • Certificates of Completion for all courses
  • New courses added every month! — stay on top of state-of-the-art trends in computer vision and deep learning

PyImageSearch University is really the best Computer Visions "Masters" Degree that I wish I had when starting out. Being able to access all of Adrian's tutorials in a single indexed page and being able to start playing around with the code without going through the nightmare of setting up everything is just amazing. 10/10 would recommend.

Sanyam BhutaniMachine Learning Engineer and 2x Kaggle Master
Close