search

yaoxufeng / PCL-Proxy-based-Contrastive-Learning-for-Domain-Generalization

MIT License
57 stars 6 forks source link

how to set the parameters #5

Closed Cassiatora closed 2 years ago

Cassiatora commented 2 years ago

Hi, I want to run the algorithms with datasets of another task. But the results are not good with the same parameters as PACS. Could you please share some experience with setting the parameters? This is the log. Thank you. log.txt

image

yaoxufeng commented 2 years ago

Hi Cassiatora,

Thanks again for your interest in our work. You can try different scale factor, hidden-size of the encoder and learning rate schedule for encoder and fea_proj.

Cassiatora commented 2 years ago

Hi Cassiatora,

Thanks again for your interest in our work. You can try different scale factor, hidden-size of the encoder and learning rate schedule for encoder and fea_proj.

Thanks for your quick reply. I don't know the specific meaning of tr_outloss、env0_out、env1_in. Would you mind telling me? Thank you for your patience in replying！

luffy-lxd commented 10 months ago

Hi Cassiatora,

Thanks again for your interest in our work. You can try different scale factor, hidden-size of the encoder and learning rate schedule for encoder and fea_proj.

hi，could you please share your parameters of TerraIncognita datastet? i have tried many times ,but the results are not good with the same parameters as TerraIncognita. thanks a lot!