Results are not getting reproduced of Office31

[shreejalt]

Hi @NaJaeMin92, I tried using the repo but I was unable to reproduce the results office dataset. My SDM accuracy got decreased to 51% and TDM is stagnant at 6%.

[NaJaeMin92]

Hi @shreejalt

As I mentioned in https://github.com/NaJaeMin92/FixBi/issues/9#issue, please use baseline weights such as DANN, MSTN and check your hyper-parameters according to the performance of the baseline models.

[Bang2hen1iu]

question. I follow #9, but neither source-only pretrained weight nor DANN pretrained weight gets good result. Is pretrained weight means the weight of encoder? Or perhaps the whole baseline architecture should be trained together with fixbi?

[codezjt]

Hi @NaJaeMin92, I tried using the repo but I was unable to reproduce the results office dataset. My SDM accuracy got decreased to 51% and TDM is stagnant at 6%.

Have you resolve the question? I got the same result.

[WANGBohaO-jpg]

I got the same result.