Load not exactly the same number of weights

[JoeHEZHAO]

Hello

Thanks for releasing your work. Big help for someone prefer pytorch like me.

When I check the test_model.py code, one thing confused me. I found out that the BNInception has total 485 components. While the ucf101_rgb.pth only recover 416 of them.

len(list(net.state_dict()) == 485 len(list(base_dict)) == 416

Is this normal ? Does this affect total accuracy ?

Best

[JoeHEZHAO]

[JoeHEZHAO]

sry I closed it by mistake.

I upload the 69 un-recovered weights name. Seems like they all related to batch track