Looping through bounding boxes obtained through inference with tensorflow object detection api2

Question

I have used the tensorflow object detection api2 for a model , now I would want to loop through multiple bounding boxes in a single image and create a separate image for each bounding box. The code I have used, with which I am able to get a image of one bounding box but unable to loop through, I think I have misunderstood the contents of the detections file.

category_index = label_map_util.create_category_index_from_labelmap(PATH_TO_LABELS)


image_np = np.array(Image.open(test_image_path))

print('Running inference for {}... '.format(test_image_path), end='')

input_tensor = tf.convert_to_tensor(image_np)
input_tensor = input_tensor[tf.newaxis, ...]
#input_tensor = input_tensor[:, :, :, :3]
#input_tensor = tf.convert_to_tensor(np.expand_dims(image_np, 0), dtype=tf.float32)
detections = detect_fn(input_tensor)

num_detections = int(detections.pop('num_detections'))
detections = {key: value[0, :num_detections].numpy()
              for key, value in detections.items()}
detections['num_detections'] = num_detections

# detection_classes should be ints.
detections['detection_classes'] = detections['detection_classes'].astype(np.int64)


image_np_with_detections = image_np.copy()

viz_utils.visualize_boxes_and_labels_on_image_array(
        image_np_with_detections,
        detections['detection_boxes'],
        detections['detection_classes'],
        detections['detection_scores'],
        category_index,
        use_normalized_coordinates=True,
        max_boxes_to_draw=500,
        min_score_thresh=.2,
        agnostic_mode=False)
plt.figure()
plt.imshow(image_np_with_detections)
print('Done')
plt.show()

def crop_objects(image, image_np_with_detections, detections):
    global ymin, ymax, xmin, xmax
    width, height = image.size
#Coordinates of detected objects
    ymin = int(detections['detection_boxes'][0][0]*height)
    xmin = int(detections['detection_boxes'][0][1]*width)
    ymax = int(detections['detection_boxes'][0][2]*height)
    xmax = int(detections['detection_boxes'][0][3]*width)
    crop_img = image_np_with_detections[ymin:ymax, xmin:xmax]


    if detections['detection_scores'][0] < 0.5:
        crop_img.fill(0)

    #Save cropped object into image
    cv2.imwrite('D:\\pcb_project\\test_images' + '.png', crop_img)
    return ymin, ymax, xmin, xmax

Answer 1

Just looked into the detections file. From there it was simple. For anyone else who wants it, this is working for me.

def crop_objects(image, image_np_with_detections, detections, i):
#global ymin, ymax, xmin, xmax
width, height = image.size
i=i
#Coordinates of detected objects
for detection in detections['detection_boxes']:
    ymin = int(detection[0]*height)
    xmin = int(detection[1]*width)
    ymax = int(detection[2]*height)
    xmax = int(detection[3]*width)
    crop_img = image_np_with_detections[ymin:ymax, xmin:xmax]
    

    if detections['detection_scores'][i] < detection_threshold:
        crop_img.fill(0)
    
#Save cropped object into image
    cv2.imwrite('D:\\pcb_project\\test_images' +  str(i) +'.png', crop_img)
    i+=1

Answer 2

The function doesn't work for me. This code is doing the job though:

#export cropped objects/boxes
detection_threshold=.3 #confidence of detection
height = image.shape[0]
width = image.shape[1]
box_index = 0
for detection in detections['detection_boxes']:
      ymin = int(detection[0]*height)
      xmin = int(detection[1]*width)
      ymax = int(detection[2]*height)
      xmax = int(detection[3]*width)
      crop_img = image_np[ymin:ymax, xmin:xmax]
      if detections['detection_scores'][box_index] > detection_threshold:
            label = category_index[detections['detection_classes'][box_index]]['name']
            print(str(label))
            print(str(detections['detection_scores'][box_index]) + '\n-------------')
#Save cropped object into image
            file_name = your_file_name
            cv2.imwrite(your_path_to_saved_images + str(file_name) + 'object-' + str(box_index+1) +'.png', crop_img)
      box_index += 1

Answer 3

Why is it less than in the code of if detections['detection_scores'][i] < detection_threshold: ? If I am correct it should be greater-equals to, ie

if detections['detection_scores'][i] >= detection_threshold:
    # crop your image