Hao Ren

I have other GPUs, like RTX 2060 Super or TiTAN X, so I could run that in these GPUs now.

ok, thanks, I wonder why the conda env named `torch_zoo` but not `raft` for `RAFT`, it's confusing. And I also found an inconsistent problem for extracting features for [THUMOS14](http://crcv.ucf.edu/THUMOS14/download.html) dataset...

It can work, and there are another warning, just too annoying

Thank you for your patience explain

yeah, I have noticed that, and I think if we want to compute the real temporal area, for example, [1, 2], the time lasted 1 second maybe more intuitive, and...

@amm040341 I also encountered the same problem, when I train the `C3D` model with official train/test split1, the testing accuracy is so low.

@wave-transmitter It seems like a typo error, it should be on training process. And I think the function `randomflip` has a bug, in the `for` loop, it calls `cv2.flip` two...

@wave-transmitter The common solution is that Normalization should be done after the data have been scaled to [0,1], we usually call the function `ToTensor()` then follow with some Normalization ops...

@wave-transmitter I trained `C3D` with official split1 from scratch, not used the pre-trained model, and you could test the `C3D` model from scratch just change one line code in `normalize`...

@szagoruyko But I found your `conv2d_depthwise_kernel` used `$(Dtype) value` to accumulate the result, and on `test_modules` function, it uses FloatTensor to test, according to the `utils.py`, the `$(Dtype) value` will...