KaiyangZhou / Dassl.pytorch

A PyTorch toolbox for domain generalization, domain adaptation and semi-supervised learning.
MIT License
1.23k stars 171 forks source link

Dynamic Domain Generalization #43

Closed siaimes closed 2 years ago

siaimes commented 2 years ago

Dynamic Domain Generalization: https://arxiv.org/abs/2205.13913

Signed-off-by: siaimes 34199488+siaimes@users.noreply.github.com

KaiyangZhou commented 2 years ago

Seems like the changes are safe and won't affect other methods. Have you checked this carefully?

Just double check, have you run "linter.sh" to remove potential inconsistencies?

siaimes commented 2 years ago

Seems like the changes are safe and won't affect other methods. Have you checked this carefully?

Just double check, have you run "linter.sh" to remove potential inconsistencies?

  1. Yes;
  2. Yes.
KaiyangZhou commented 2 years ago

I've checked the changes. Most look good. Please use double quotes instead of single quotes.

What's new about DomainNetDG? Can you also add some config files, which can be used to reproduce the results in the paper?

siaimes commented 2 years ago

What's new about DomainNetDG?

Nothing but removing useless train_u from dassl/data/da/domainnet.py to save some memory.

Can you also add some config files, which can be used to reproduce the results in the paper?

I have found no other configuration files in Dassl, so I did not add the configuration files. Currently, the configuration files are in our own project, you can refer to our git repository: https://github.com/MetaVisionLab/DDG.

KaiyangZhou commented 2 years ago

If there is no new in DomainnetDG, please remove it. train_u won't be used in this case https://github.com/KaiyangZhou/Dassl.pytorch/blob/master/dassl/engine/trainer.py#L593

It'd be good if you can provide some configs in https://github.com/KaiyangZhou/Dassl.pytorch/tree/master/configs/trainers/dg

siaimes commented 2 years ago

It'd be good if you can provide some configs in https://github.com/KaiyangZhou/Dassl.pytorch/tree/master/configs/trainers/dg

In my opinion, it would be better for the config files to be independent of Dassl, which I have placed here: https://github.com/MetaVisionLab/DDG/tree/main/dassl, where I also provide the startup script.