Closed pageedward closed 4 years ago
Currently we don't support training a sample without label, so you should fillter them out like in coco. https://github.com/lufficc/SSD/blob/c16e4591884d467d60094c6382ed1975c541a5c7/ssd/data/datasets/coco.py#L34-L36
@lufficc no objects in a train sampler,it will only have classify loss, but the code no deal with this issue,it right?