Why model networkbased on resnet50？what if resnet101？

[swg209]

I got confused why all the networks are based on resnet50, what if based on resnet101?

[zhangzhengjie626]

When using temporal attention，map is just 70.4%. what about you?

[swg209]

@MaverickIce I got 70.2% mAP ， but I found that temporal pooling model running on single GPU with the defuault hyper parameters could achieve a result which mAP is 73.2% , rank-1 is 81.7% when running 350 epoch。