The first line in anchor_manipulator.py: gt_labels = gt_labels + (-1 * tf.to_int64(matched_gt < -1))
The second line in train_ssd.py: batch_negtive_mask = tf.equal(cls_targets, 0)#tf.logical_and(tf.equal(cls_targets, 0), match_scores > 0.)
In my understanding, cls_targets is batched gt_labels, each gt_label is:
=1: object
=0: background including all negatives considered as background
-1: for ignores
But in the second line: cls_target is compared with 0. So what about -1 for ignore???
Didn't you handle that case???
The first line in anchor_manipulator.py: gt_labels = gt_labels + (-1 * tf.to_int64(matched_gt < -1)) The second line in train_ssd.py: batch_negtive_mask = tf.equal(cls_targets, 0)#tf.logical_and(tf.equal(cls_targets, 0), match_scores > 0.) In my understanding, cls_targets is batched gt_labels, each gt_label is: