YOLO.pytorch
YOLO v1
I remove the last two fc layers and use a conv layer to instead. It is hard to train fc layers, i do not have time to reproduct yolo like original paper.
This code just has two important part, network used resnet50 like a backbone and yolo loss. I just train code on Voc2007 for a verification. If you want use this code you should modify your Voc2007 dataset location in train code.
Question
I do not understand why used two bbox to pridict object. The two bbox will be similar because loss critize them equally. Need your help!!! Thank you very much!!!