HuaZheLei

problems about model saving

1

Hi, thanks for your excellent work. You use 'DataParallel' to support multi-gpu training. The standard way for saving models with 'DataParallel' is to use `torch.save(model.module.state_dict(), PATH)`. However, in your code,...

your emd loss seems wrong

1

Your implement is as below: > samplewise_emd = torch.sqrt(torch.mean(torch.pow(torch.abs(cdf_diff), 2))) > return samplewise_emd.mean() However, I think the correct implement should be: > emd = torch.sqrt(torch.mean(torch.pow(torch.abs(cdf_diff), 2), -1)) > return emd.mean()...

About Ranking Pair Generation!

3

Hi, thanks for your great job. I want to know how you generate ranking pair according to your annotaions. In the paper, you said > We first generate ranking pairs...

Does anyone succeed on imagenet?

5

I tried some small Ps, some small Ks and many learning rates. But I always get a loss of 0.693. Anyone can share his/her experience on imagenet?

[Bug] memory leak

7

**Description:** When I use a 'for' loop to cut a sequence of videos, the memory raise all the time. **Command:** for video_path in video_lists: video = open_video(video_path) scene_manager = SceneManager()...