How to count objects in Tensorflow Object Detection API

Question

I am executing https://github.com/tensorflow/tensorflow this example of detecting objects in image.

I want to get count of detected objects following is the code that gives me detected object drawn in an image. But I am not able to get count of detected objects.

with detection_graph.as_default():
with tf.Session(graph=detection_graph) as sess:
    for image_path in TEST_IMAGE_PATHS:
      image = Image.open(image_path)
      # the array based representation of the image will be used later in order to prepare the
      # result image with boxes and labels on it.
      image_np = load_image_into_numpy_array(image)
      # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
      image_np_expanded = np.expand_dims(image_np, axis=0)
      image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
      # Each box represents a part of the image where a particular object was detected.
      boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
      # Each score represent how level of confidence for each of the objects.
      # Score is shown on the result image, together with the class label.
      scores = detection_graph.get_tensor_by_name('detection_scores:0')
      classes = detection_graph.get_tensor_by_name('detection_classes:0')
      num_detections = detection_graph.get_tensor_by_name('num_detections:0')
      # Actual detection.
      (boxes, scores, classes, num_detections) = sess.run(
          [boxes, scores, classes, num_detections],
          feed_dict={image_tensor: image_np_expanded})
      # Visualization of the results of a detection.
      vis_util.visualize_boxes_and_labels_on_image_array(
          image_np,
          np.squeeze(boxes),
          np.squeeze(classes).astype(np.int32),
          np.squeeze(scores),
          category_index,
          use_normalized_coordinates=True,
          line_thickness=1)
      plt.figure(figsize=IMAGE_SIZE)
      plt.imshow(image_np)

This is the block of code that gives actual object detection shown in below image:

How can I get the object count?

this gives me information of those classes. I need a count which I got from printing length of boxes.shape. Any ways thank you. — Rohit Salunke
– Rohit Salunke, Commented Aug 7, 2017 at 10:10
Here is an Object Counting API for TensorFlow: github.com/ahmetozlu/tensorflow_object_counting_api , you can use it to count objects and more for developing object detection and counting based projects.. — Ozlu
– Ozlu, Commented Jul 3, 2018 at 21:21

Ozlu · Accepted Answer · 2021-11-20 09:43:43Z

7

You can use the TensorFlow Object Counting API that is an open source framework built on top of TensorFlow that makes it easy to develop object counting systems to count any objects! Moreover, it provides sample projects so you can adopt them to develop your own specific case studies!

Sample Project#1 is "Pedestrian Counting":

Sample Project#2 is "Vehicle Counting":

Sample Project#3 is "Object Counting in Real-Time":

See the TensorFlow Object Counting API for more info and please give a star that repo for showing your support to open source community if you find it useful!

edited Nov 20, 2021 at 9:43

answered Aug 1, 2018 at 20:09

Ozlu

1,3051 gold badge14 silver badges20 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

Rohit Salunke · Accepted Answer · 2018-07-06 03:32:43Z

4

Solve it simply print length of boxes.shape

print(len(boxes.shape))

edited Jul 6, 2018 at 3:32

answered Aug 7, 2017 at 10:08

Rohit Salunke

1,1031 gold badge10 silver badges15 bronze badges

Comments

Eric Aya · Accepted Answer · 2017-10-02 13:52:34Z

2

You should check scores and count objects as manual. Code is here:

#code to test image start

    (boxes, scores, classes, num) = sess.run(
        [detection_boxes, detection_scores, detection_classes, num_detections],
        feed_dict={image_tensor: image_np_expanded})

#code to test image finish

#add this part to count objects

    final_score = np.squeeze(scores)    
        count = 0
        for i in range(100):
            if scores is None or final_score[i] > 0.5:
                    count = count + 1

#count is the number of objects detected

edited Oct 2, 2017 at 13:52

Eric Aya

70.2k36 gold badges190 silver badges266 bronze badges

answered Oct 2, 2017 at 13:49

Hakan A.

211 bronze badge

1 Comment

Rashi Over a year ago

if scores is None , does it mean it contain an object ?

rumdrums · Accepted Answer · 2017-08-19 19:05:06Z

1

It's important to note that the number of boxes is always 100.

If you look at the code that actually draws the boxes, i.e., the vis_util.visualize_boxes_and_labels_on_image_array function, you'll see that they're defining a threshold -- min_score_thresh=.5 -- to limit the boxes drawn to only those detections in which the score is > 0.5. You can think of this as only drawing boxes where the probability of accurate detection is >50%. You can adjust this threshold up or down to increase the number of boxes drawn. If you decrease it too low, however, you will get a lot of inaccurate boxes.

edited Aug 19, 2017 at 19:05

answered Aug 19, 2017 at 18:54

rumdrums

1,3922 gold badges13 silver badges27 bronze badges

1 Comment

Paul Over a year ago

Just a note: If you want to know if your object detector has confidently detected any object before visualizing the image, you can put tf.reduce_max(detections['detection_scores']) > 0.8 in an "if" statement. The code returns True if the detector has a detection of over 80% confidence.

CertainPerformance · Accepted Answer · 2018-07-12 05:00:13Z

1

add this part to count objects

final_score = np.squeeze(scores)    
    count = 0
    for i in range(100):
        if scores is None or final_score[i] > 0.5:
                count = count + 1

count is the number of objects detected

this part will print count but will print it in continuous manner can it be used to print only once like final count = some value instead of printing it repeatedly

edited Jul 12, 2018 at 5:00

CertainPerformance

373k55 gold badges354 silver badges359 bronze badges

answered Jul 12, 2018 at 4:28

Shivam

1651 silver badge9 bronze badges

Collectives™ on Stack Overflow

How to count objects in Tensorflow Object Detection API

5 Answers 5

Comments

Comments

1 Comment

1 Comment

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

5 Answers 5

Comments

Comments

1 Comment

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related