Intersection over Union (IoU) for object detection

Figure 1: An example of detecting a stop sign in an image. The predicted bounding box is drawn in red while the ground-truth bounding box is drawn in green. Our goal is to compute the Intersection of Union between these bounding box.

Today’s blog post is inspired from an email I received from Jason, a student at the University of Rochester.

Jason is interested in building a custom object detector using the HOG + Linear SVM framework for his final year project. He understands the steps required to build the object detector well enough — but he isn’t sure how to evaluate the accuracy of his detector once it’s trained.

His professor mentioned that he should use the Intersection over Union (IoU) method for evaluation, but Jason’s not sure how to implement it.

I helped Jason out over email by:

  1. Describing what Intersection over Union is.
  2. Explaining why we use Intersection over Union to evaluate object detectors.
  3. Providing him with some example Python code from my own personal library to perform Intersection over Union on bounding boxes.

My email really helped Jason finish getting his final year project together and I’m sure he’s going to pass with flying colors.

With that in mind, I’ve decided to turn my response to Jason into an actual blog post in hopes that it will help you as well.

To learn how to evaluate your own custom object detectors using the Intersection over Union evaluation metric, just keep reading.

Looking for the source code to this post?
Jump right to the downloads section.

Intersection over Union for object detection

In the remainder of this blog post I’ll explain what the Intersection over Union evaluation metric is and why we use it.

I’ll also provide a Python implementation of Intersection over Union that you can use when evaluating your own custom object detectors.

Finally, we’ll look at some actual results of applying the Intersection over Union evaluation metric to a set of ground-truth and predicted bounding boxes.

What is Intersection over Union?

Intersection over Union is an evaluation metric used to measure the accuracy of an object detector on a particular dataset. We often see this evaluation metric used in object detection challenges such as the popular PASCAL VOC challenge.

You’ll typically find Intersection over Union used to evaluate the performance of HOG + Linear SVM object detectors and Convolutional Neural Network detectors (R-CNN, Faster R-CNN, YOLO, etc.); however, keep in mind that the actual algorithm used to generate the predictions doesn’t matter.

Intersection over Union is simply an evaluation metric. Any algorithm that provides predicted bounding boxes as output can be evaluated using IoU.

More formally, in order to apply Intersection over Union to evaluate an (arbitrary) object detector we need:

  1. The ground-truth bounding boxes (i.e., the hand labeled bounding boxes from the testing set that specify where in the image our object is).
  2. The predicted bounding boxes from our model.

As long as we have these two sets of bounding boxes we can apply Intersection over Union.

Below I have included a visual example of a ground-truth bounding box versus a predicted bounding box:

Figure 1: An example of detecting a stop sign in an image. The predicted bounding box is drawn in red while the ground-truth bounding box is drawn in green. Our goal is to compute the Intersection of Union between these bounding box.

Figure 1: An example of detecting a stop sign in an image. The predicted bounding box is drawn in red while the ground-truth bounding box is drawn in green. Our goal is to compute the Intersection of Union between these bounding box.

In the figure above we can see that our object detector has detected the presence of a stop sign in an image.

The predicted bounding box is drawn in red while the ground-truth (i.e., hand labeled) bounding box is drawn in green.

Computing Intersection over Union can therefore be determined via:

Figure 2: Computing the Intersection of Union is as simple as dividing the area of overlap between the bounding boxes by the area of union (thank you to the excellent Pittsburg HW4 assignment for the inspiration for this figure).

Figure 2: Computing the Intersection of Union is as simple as dividing the area of overlap between the bounding boxes by the area of union (thank you to the excellent Pittsburg HW4 assignment for the inspiration for this figure).

Examining this equation you can see that Intersection over Union is simply a ratio.

In the numerator we compute the area of overlap between the predicted bounding box and the ground-truth bounding box.

The denominator is the area of union, or more simply, the area encompassed by both the predicted bounding box and the ground-truth bounding box.

Dividing the area of overlap by the area of union yields our final score — the Intersection over Union.

Where are you getting the ground-truth examples from?

Before we get too far, you might be wondering where the ground-truth examples come from. I’ve mentioned before that these images are “hand labeled”, but what exactly does that mean?

You see, when training your own object detector (such as the HOG + Linear SVM method), you need a dataset. This dataset should be broken into (at least) two groups:

  1. training set used for training your object detector.
  2. testing set for evaluating your object detector.

You may also have a validation set used to tune the hyperparameters of your model.

Both the training and testing set will consist of:

  1. The actual images themselves.
  2. The bounding boxes associated with the object(s) in the image. The bounding boxes are simply the (x, y)-coordinates of the object in the image.

The bounding boxes for the training and testing sets are hand labeled and hence why we call them the “ground-truth”.

Your goal is to take the training images + bounding boxes, construct an object detector, and then evaluate its performance on the testing set.

An Intersection over Union score > 0.5 is normally considered a “good” prediction. 

Why do we use Intersection over Union?

If you have performed any previous machine learning in your career, specifically classification, you’ll likely be used to predicting class labels where your model outputs a single label that is either correct or incorrect.

This type of binary classification makes computing accuracy straightforward; however, for object detection it’s not so simple.

In all reality, it’s extremely unlikely that the (x, y)-coordinates of our predicted bounding box are going to exactly match the (x, y)-coordinates of the ground-truth bounding box.

Due to varying parameters of our model (image pyramid scale, sliding window size, feature extraction method, etc.), a complete and total match between predicted and ground-truth bounding boxes is simply unrealistic.

Because of this, we need to define an evaluation metric that rewards predicted bounding boxes for heavily overlapping with the ground-truth:

Figure 3: An example of computing Intersection over Unions for various bounding boxes.

Figure 3: An example of computing Intersection over Unions for various bounding boxes.

In the above figure I have included examples of good and bad Intersection over Union scores.

As you can see, predicted bounding boxes that heavily overlap with the ground-truth bounding boxes have higher scores than those with less overlap. This makes Intersection over Union an excellent metric for evaluating custom object detectors.

We aren’t concerned with an exact match of (x, y)-coordinates, but we do want to ensure that our predicted bounding boxes match as closely as possible — Intersection over Union is able to take this into account.

Implementing Intersection over Union in Python

Now that we understand what Intersection over Union is and why we use it to evaluate object detection models, let’s go ahead and implement it in Python.

Before we get started writing any code though, I want to provide the five example images we will be working with:

Figure 4: In this example, we'll be detecting the presence of cars in images.

Figure 4: In this example, we’ll be detecting the presence of cars in images.

These images are part of the CALTECH-101 dataset used for both image classification and object detection.

Inside the PyImageSearch Gurus course I demonstrate how to train a custom object detector to detect the presence of cars in images like the ones above using the HOG + Linear SVM framework.

I have provided a visualization of the ground-truth bounding boxes (green) along with the predicted bounding boxes (red) from the custom object detector below:

Figure 5: Our goal is to evaluate the performs of our object detector by using Intersection of Union. Specifically, we want to measure the accuracy of the predicted bounding box (red) against the ground-truth (green).

Figure 5: Our goal is to evaluate the performance of our object detector by using Intersection of Union. Specifically, we want to measure the accuracy of the predicted bounding box (red) against the ground-truth (green).

Given these bounding boxes, our task is to define the Intersection over Union metric that can be used to evaluate how “good (or bad) our predictions are.

With that said, open up a new file, name it intersection_over_union.py , and let’s get coding:

We start off by importing our required Python packages. We then define a Detection  object that will store three attributes:

  • image_path : The path to our input image that resides on disk.
  • gt : The ground-truth bounding box.
  • pred : The predicted bounding box from our model.

As we’ll see later in this example, I’ve already obtained the predicted bounding boxes from our five respective images and hardcoded them into this script to keep the example short and concise.

For a complete review of the HOG + Linear SVM object detection framework, please refer to this blog post. And if you’re interested in learning more about training your own custom object detectors from scratch, be sure to check out the PyImageSearch Gurus course.

Let’s go ahead and define the bb_intersection_over_union  function, which as the name suggests, is responsible for computing the Intersection over Union between two bounding boxes:

This method requires two parameters: boxA  and boxB , which are presumed to be our ground-truth and predicted bounding boxes (the actual order in which these parameters are supplied to bb_intersection_over_union  doesn’t matter).

Lines 11-14 determine the (x, y)-coordinates of the intersection rectangle which we then use to compute the area of the intersection (Line 17).

The interArea  variable now represents the numerator in the Intersection over Union calculation.

To compute the denominator we first need to derive the area of both the predicted bounding box and the ground-truth bounding box (Lines 21 and 22).

The Intersection over Union can then be computed on Line 27 by dividing the intersection area by the union area of the two bounding boxes, taking care to subtract out the intersection area from the denominator (otherwise the intersection area would be doubly counted).

Finally, the Intersection over Union score is returned to the calling function on Line 30.

Now that our Intersection over Union method is finished, we need to define the ground-truth and predicted bounding box coordinates for our five example images:

As I mentioned above, in order to keep this example short(er) and concise, I have manually obtained the predicted bounding box coordinates from my HOG + Linear SVM detector. These predicted bounding boxes (And corresponding ground-truth bounding boxes) are then hardcoded into this script.

For more information on how I trained this exact object detector, please refer to the PyImageSearch Gurus course.

We are now ready to evaluate our predictions:

On Line 41 we start looping over each of our examples  (which are Detection  objects).

For each of them, we load the respective image  from disk on Line 43 and then draw the ground-truth bounding box in green (Lines 47 and 48) followed by the predicted bounding box in red (Lines 49 and 50).

The actual Intersection over Union metric is computed on Line 53 by passing in the ground-truth and predicted bounding box.

We then write the Intersection over Union value on the image  itself followed by our console as well.

Finally, the output image is displayed to our screen on Lines 59 and 60.

Comparing predicted detections to the ground-truth with Intersection over Union

To see the Intersection over Union metric in action, make sure you have downloaded the source code + example images to this blog post by using the “Downloads” section found at the bottom of this tutorial.

After unzipping the archive, execute the following command:

Our first example image has an Intersection over Union score of 0.7980, indicating that there is significant overlap between the two bounding boxes:

Figure 6: Computing the Intersection of Union using Python.

Figure 6: Computing the Intersection of Union using Python.

The same is true for the following image which has an Intersection over Union score of 0.7899:

Figure 7: A slightly better Intersection over Union score.

Figure 7: A slightly better Intersection over Union score.

Notice how the ground-truth bounding box (green) is wider than the predicted bounding box (red). This is because our object detector is defined using the HOG + Linear SVM framework which requires us to specify a fixed size sliding window (not to mention, an image pyramid scale and the HOG parameters themselves).

Ground-truth bounding boxes will naturally have a slightly different aspect ratio than the predicted bounding boxes, but that’s okay provided that the Intersection over Union score is > 0.5 — as we can see, this still a great prediction.

The next example demonstrates a slightly “less good” prediction where our predicted bounding box is much less “tight” than the ground-truth bounding box:

Figure 8: Deriving the Intersection of Union evaluation metric for object detection.

Figure 8: Deriving the Intersection of Union evaluation metric for object detection.

The reason for this is because our HOG + Linear SVM detector likely couldn’t “find” the car in the lower layers of the image pyramid and instead fired near the top of the pyramid where the image is much smaller.

The following example is an extremely good detection with an Intersection over Union score of 0.9472:

Figure 9: Measuring object detection performance using Intersection over Union.

Figure 9: Measuring object detection performance using Intersection over Union.

Notice how the predicted bounding box nearly perfectly overlaps with the ground-truth bounding box.

Here is one final example of computing Intersection over Union:

Figure 10: Intersection over Union for evaluating object detection algorithms.

Figure 10: Intersection over Union for evaluating object detection algorithms.

Want to train your own custom object detectors?

If you enjoyed this tutorial and want to learn more about training your own custom object detectors, you’ll definitely want to take a look at the PyImageSearch Gurus course — the most complete, comprehensive computer vision course online today.

Inside the course, you’ll find over 168 lessons covering 2,161+ pages of content on Object Detection, Image Classification, Convolutional Neural Networks, and much more.

To learn more about the PyImageSearch Gurus course (and grab your FREE sample lessons + course syllabus), just click the button below:

Click here to learn more about PyImageSearch Gurus!

Summary

In this blog post I discussed the Intersection over Union metric used to evaluate object detectors. This metric can be used to assess any object detector provided that (1) the model produces predicted (x, y)-coordinates [i.e., the bounding boxes] for the object(s) in the image and (2) you have the ground-truth bounding boxes for your dataset.

Typically, you’ll see this metric used for evaluating HOG + Linear SVM and CNN-based object detectors.

To learn more about training your own custom object detectors, please refer to this blog post on the HOG + Linear SVM framework along with the PyImageSearch Gurus course where I demonstrate how to implement custom object detectors from scratch.

Finally, before you go, be sure to enter your email address in the form below to be notified when future PyImageSearch blog posts are published — you won’t want to miss them!

Downloads:

If you would like to download the code and images used in this post, please enter your email address in the form below. Not only will you get a .zip of the code, I’ll also send you a FREE 11-page Resource Guide on Computer Vision and Image Search Engines, including exclusive techniques that I don’t post on this blog! Sound good? If so, enter your email address and I’ll send you the code immediately!

, , , ,

29 Responses to Intersection over Union (IoU) for object detection

  1. wajih November 7, 2016 at 10:33 am #

    I was translating a code, was wondering of IoU, and now really I OWE YOU ONE 🙂 Thanks for the explanation. Helped me finish the translation in a breeze 🙂

    • Adrian Rosebrock November 7, 2016 at 2:42 pm #

      Awesome, I’m happy I could help Wajih! 🙂

  2. Walid Ahmed November 7, 2016 at 11:23 am #

    Thanks a lot.

    This was really in time for me
    however, I am still clueless how to build a classifier for object detection.
    I already built my classifier for object classification using CNN.
    Any advice?

    • Adrian Rosebrock November 7, 2016 at 2:42 pm #

      Hey Walid — I would suggest starting by reading about the HOG + Linear SVM detector, a classic method used for object detection. I also demonstrate how to implement this system inside the PyImageSearch Gurus course. As for using a CNN for object detection, I will be covering that in my next book.

  3. kiran November 7, 2016 at 9:04 pm #

    Awesome. You made it pretty simple to understand. I request you to post a similar blog on evaluation metrics for object segmentation tasks as well. Thank you.

    • Adrian Rosebrock November 10, 2016 at 8:50 am #

      Most of my work focuses on object detection rather than pixel-wise segmentations of an image but I’ll certainly consider it for the future.

  4. jsky November 7, 2016 at 10:43 pm #

    I implemented this in my car detection framework for my FYP too.
    Its called the Jaccard Index, and is a standard measure for evaluating such record retrieval.
    https://en.wikipedia.org/wiki/Jaccard_index

  5. abby November 10, 2016 at 8:54 am #

    thank you for the tutorial

    • Adrian Rosebrock November 10, 2016 at 9:40 am #

      No problem, happy I can help!

  6. Aamer November 12, 2016 at 2:33 am #

    what if our hog+svm model predicts multiple bounding boxes.

    In that case, will we iterate over all such predicted bounding boxes and see for the one which gets the max value for the Intersection/Union ratio ?

    • Adrian Rosebrock November 14, 2016 at 12:12 pm #

      If you detector predicts multiple bounding boxes for the same object then you should be applying non-maxima suppression. If you are predicting bounding boxes for multiple objects then you should be computing IoU for each of them.

  7. Miej November 28, 2016 at 10:47 am #

    This code fails. Specifically, it gets broken when comparing two non-overlapping bounding boxes by providing a non-negative value for interArea when the boxes can be separated into diagonally opposing quadrants by a vertical and a horizontal line. This could be easily remedied with a simple catch for such cases.

  8. Miej November 28, 2016 at 11:14 am #

    that said, it’s still quite handy. thanks!

  9. auro November 28, 2016 at 2:34 pm #

    Do we need to consider the case where the two boxes to not intersect at all?
    Look up the MATLAB code at https://github.com/rbgirshick/voc-dpm/blob/master/utils/boxoverlap.m

    • Adrian Rosebrock November 28, 2016 at 2:40 pm #

      Good point, thank you for pointing this out Auro.

  10. Rimphy Darmanegara December 8, 2016 at 11:55 pm #

    So, in case of negative result just return zero?

    Thank you for the code.

    • Adrian Rosebrock December 10, 2016 at 7:14 am #

      Correct. In case of negative (non-overlapping) objects the return value would be zero.

  11. Anivini December 21, 2016 at 4:34 am #

    How to give bounding box parameters from line 33 to 38?

  12. Jere December 22, 2016 at 1:34 pm #

    This is helpful knowledge. You have made me understand about the method IoU used in fast rcnn. Thanks you are one of the best people in explaining concepts easily

    • Adrian Rosebrock December 23, 2016 at 10:52 am #

      Thank you for the kind words Jere 🙂

      • sabrine December 29, 2016 at 8:54 am #

        Good evening, I worked with HOG detector and I used Matlab software, I want to calculate the rate of overlap between the detection of the ground truth and my detection file, please help me.

        • Adrian Rosebrock December 31, 2016 at 1:26 pm #

          If you have the predictions from your MATLAB detector you can either (1) write them to file and use my Python code to read them and compare them to the ground-truth or (2) implement this function in MATLAB.

          • sabrine January 1, 2017 at 6:04 am #

            OK thanks
            I will try to implement on matlab
            Because I do not know how to use python

  13. Anivini January 2, 2017 at 4:14 am #

    I am confused about detection.gt[:2] and detection.gt[2:] in lines 47 to 50. What is actually specifed by :2 and 2: . I surfed but couldn’t get an answer.

  14. goingmyway January 15, 2017 at 4:56 am #

    Awesome post. A clear explanation of IoU.

    • Adrian Rosebrock January 15, 2017 at 12:03 pm #

      Thank you, I’m happy to hear you enjoyed it! 🙂

  15. secret February 22, 2017 at 12:00 am #

    Hey I would like to know how to compute the repeatability factor ,corresponding count and recall and precision to evaluate feature detector in python

    I would like to know is there a function in python similar to the one in C++ if not then how do I proceed

Leave a Reply