tntek / source-free-domain-adaptation Goto Github PK

Hello, thanks for your great work. While I was able to reproduce the results for B-32, but found that it is hard to reproduce the results for RN.

I modified the ARCH : ViT-B/32 to RN50 in the config file for the RN experiment, and achieved a result of 50.56% for the office-home A->C task (paper : 62.6%).

In particular, in target domains such as office31-amazon and office-home-clipart, the accuracy of the zero-shot CLIP model was measured about 20%...

I wonder if I miss something?
I appreciate your help.

I tried to create the conda virtual environment using the .yml file, but it says it cannot find some, so I deleted them, then the error was gone, but it stuck and installed nothing from the .yml file.

Hi, thanks for your great work.
I have a question after reading the paper and the code.
It seems that the loss L_MCE is computed at lines 180-184 in src/methods/net/difo.py,
but it does not exactly match the form in the paper.
Could you give me some hints? Thanks for your reply.

Thanks for your great work and sharing.

I noticed that in some paper, the data_list is arranged in order, 1,2,3, and in your data_list like Art_list.txt, it is a random order.
When you report the results in the table, do you use the same style list file for all the methods? I notice that these two style list can produce difference results with a gap.

Did the reported results in the paper use the accuracy from the final Epoch or the Best Epoch? I noticed that sometimes the final epoch is not the best one.

Thank you very much.