Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Questions related to the yaml file #1

Open
batman47steam opened this issue Jul 12, 2023 · 5 comments
Open

Questions related to the yaml file #1

batman47steam opened this issue Jul 12, 2023 · 5 comments

Comments

@batman47steam
Copy link
Contributor

batman47steam commented Jul 12, 2023

The default yaml file seems not work.

@AgainstEntropy
Copy link
Owner

Sorry for the deprecated YAML configuration file 😿
Just remove the total_len key and change the noise_factor key to noisy.

As for the problem with model_name, I will merge your PR as a fast solution~ Thanks for your contribution 😄

@AgainstEntropy
Copy link
Owner

Hi, thank you for the response and for sharing the code and dataset. I think this work is pretty cool. I found that noise_factor also appeared once in the trainer.py, but it doesn't seem to be used anywhere else. since I haven't read the code thoroughly, I am not sure is that ok to also remove the noise_factor in log_train_cfg ? image

Thanks! 😆 It means a lot to me~

Feel free to modify the log_train_cfg dict since it is just some configuration logged to wandb for experiment tracking.

I will keep this issue open until no more minor mistake is found~

@AgainstEntropy
Copy link
Owner

Hi, the email address in the paper seems unable to reach, can you provide another email address. So that I can further consult some details.

Sure! Please contact me (the first author) at yihao.w@nyu.edu or yihaowang18@fudan.edu.cn

Sorry for the invalid email address due to departure from Shanghai AI Lab. If you would like to contact any of the other authors, feel free to send an email at the address shown on the paper, as they are still in Shanghai AI Lab.

@batman47steam
Copy link
Contributor Author

In compute_track_metrics functions, all these metrics have been averaged on the sequence length T (pcm / T, area / T, dtw / T). but in the compute_batch_metrics, it seems that the result is not averaged over batch dimensions, it returns the sum of these metrics in a batch. I'm not sure if this is why larger batch sizes lead to higher values for these metrics. I also trained with batch_size = 6, the pcm, area, and dtw metrics is much lower than batch_size = 32
image
image

@batman47steam
Copy link
Contributor Author

Hi, In the simulation dataset, for example, 944, there are some negative values in video_128.npy and video_128_noisy.npy, was an additional data enhancement operation performed?
image

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants