TensorFlow Object Detection API print objects found on image to console

Question

I\'m trying to return list of objects that have been found at image with TF Object Detection API.

To do that I\'m using print([category_index.get(i) for

Answer 1

// this will load the labels and categories along with category index

label_map = label_map_util.load_labelmap(PATH_TO_LABELS)

categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)

category_index = label_map_util.create_category_index(categories)


//to print the identified object do the following :

print category instead of category index. The index holds the numeric value and the category contains the name of the objects. Once identified with the mentioned threshold the

min_score_thresh = 0.5

print([category.get(1)] for i in classes[0] if scores[0, i] > min_score_thresh)

this will print the identified category.

Answer 2

open visualization_utils.py and add--> print(class_name) after

else:        
 class_name = 'N/A'
      display_str = '{}: {}%'.format(
          class_name,
          int(100*scores[i])) 

this will print the detected objects

Answer 3

From the function signature visualize_boxes_and_labels_on_image_array, you have to set the arguments max_boxes_to_draw, min_score_thresh,

visualize_boxes_and_labels_on_image_array(image,
                                          boxes,
                                          classes,
                                          scores,
                                          category_index,
                                          instance_masks=None,
                                          keypoints=None,
                                          use_normalized_coordinates=False,
                                          max_boxes_to_draw=20,
                                          min_score_thresh=.5,
                                          agnostic_mode=False,
                                          line_thickness=4)

Answer 4

Try to set the min_score_thresh to 0. Then you will probably see 300 detections.

Answer 5

adding print(class_name) after

else:        
 class_name = 'N/A'
      display_str = '{}: {}%'.format(
          class_name,
          int(100*scores[i]))

in visualization_utils.py file prints the detected object. I wonder where to add print command to print timestamps as well as percentage of accuracy in output.

Answer 6

As far as I can see you have 300 detections. visualize_boxes_and_labels_on_image_array shows very few of them because min_score_thresh=.5 (this is the default value) is too high for the most of them.

If you want to add such filtering to the output you can write:

min_score_thresh = 0.5
print([category_index.get(i) for i in classes[0] if scores[0, i] > min_score_thresh)

You can change min_score_thresh to choose threshold value you need. It may be useful to print the score values with the category names.