zzh8829/yolov3-tf2

What does tf.pad do?

dhwgithub opened this issue · 3 comments

dhwgithub commented

Source code link: https://github.com/zzh8829/yolov3-tf2/blob/master/yolov3_tf2/dataset.py#L115

    paddings = [[0, FLAGS.yolo_max_boxes - tf.shape(y_train)[0]], [0, 0]]
    y_train = tf.pad(y_train, paddings)

In addition, FLAGS.yolo_max_boxes I did not find the assignment statement

Can someone help me answer these two questions? Thank you

KacperPaszkowski commented

Label shape must match network output shape. Network output shape is defined by yolo_max_boxes. So if image have only 2 label and yolo_max_boxes equals 100, you need to add padding with zeros (e.g. 98 "zeros" labels) so shapes matches.

dhwgithub commented

Where is yolo_max_boxes defined or initialized? thanks