Open rogercw opened 5 years ago
Thanks for sharing the codes. I am little confused about the way to generate the ground truth labels. Hope you could help me point out what I might misunderstand. Thanks.
When generating training labels in https://github.com/leeyoshinari/YOLO_v2/blob/aa9bf78ecf461a1a3924e466024f8417d23197aa/pascal_voc.py#L61-L76, won't the labels be overwritten if two objects in the same cell? If that is the case, does that mean the first object can not be used for training?
YOLOv2假定了每个cell至多含有一个grounth truth,而在实际上基本不会出现多于1个的情况。
Thanks for sharing the codes. I am little confused about the way to generate the ground truth labels. Hope you could help me point out what I might misunderstand. Thanks.
When generating training labels in https://github.com/leeyoshinari/YOLO_v2/blob/aa9bf78ecf461a1a3924e466024f8417d23197aa/pascal_voc.py#L61-L76, won't the labels be overwritten if two objects in the same cell? If that is the case, does that mean the first object can not be used for training?