训练集trainval_GT和trainval_NEG区别是啥

[whqwill]

训练集trainval_GT和trainval_NEG区别是啥

_{Sent from PPHub}

[hwfan]

Explanation of the two sets:
GT: ground-truth(positive, with interaction labels) samples
NEG: negative(no interaction) samples

[whqwill]

OK. But I have a question that if for this model we don't use the object information, I think some negative pairs information may be same as GT pairs information in the same picture sometimes because it only use human box. So I think this problem may
affect the training results.

[hwfan]

Thanks for your suggestions, we will carefully check the human-only data in the next version to avoid this phenomenon. Pull Requests are also welcomed to improve the quality of this project :)