Kunchang Li
Kunchang Li
I want to know how to use it, too!!! Thank you very much.
加一，我也遇到了...用以前创建过腾讯云项目的APPID能正常登陆，新注册的APPID不行，很玄
In my single machine, I just simply remove comments and use `InplaceShift`, and it works. I think it may be wrong in multi-machine...
I think you can rewrite `resnet.py` and store features in `forward()` like `self.features = x`. And you can store them while runing codes.
Could you please let us know which dataset you use? I have tried to train in Sth-Sth v1 dataset, and the `val_videofolder.txt` is got by run `gen_label_sthv1.py`, which uses `validation.csv`....
For Kinetics, you can download it in [cvfoundation](https://github.com/cvdfoundation/kinetics-dataset). By the way, welcome to play with our [UniFormer](https://github.com/Sense-X/UniFormer/tree/main/video_classification), which is a new SOTA backbone for video classification.
I think you can try to set `fold_div=8/4` as user does (as the numbers of channel can be usually divided by 8/4), and you can train uni-direction TSM as follows...
I just use the code for data processing in your repo...
It is used to make sure that the number channel can be divide by 8（for example）. You can see it in MobileNetV2.
> > It is used to make sure that the number channel can be divide by 8（for example）. You can see it in MobileNetV2. > > 哪里说过 https://github.com/d-li14/mobilenetv2.pytorch/blob/master/models/imagenet/mobilenetv2.py ```python This...