YanchaoYang / FDA

Fourier Domain Adaptation for Semantic Segmentation
491 stars 79 forks source link

Question about second SST round #12

Closed LeonBre closed 3 years ago

LeonBre commented 3 years ago

Hello and thank you for your great work :-) I have a question about the performance boost of around 1,5% after the second SST round. How do you achieve that? I tried it out twice and both times my model has a worse performance than after the first SST round. Do I have to change something in the settings or do I just have to retry it a couple more times.

YanchaoYang commented 3 years ago

which epoch you are choosing for validation?

LeonBre commented 3 years ago

Thank you for your fast reply! I did not touch the settings for the number of steps. The validation is done after 100000 steps. Do I have to use more steps here?

YanchaoYang commented 3 years ago

Yeah, I guess that is the problem. So during SST you may want to use early ckpts, to prevent overfitting to wrong pseudo labels.

LeonBre commented 3 years ago

Ok I will try that.

Thanks :+1:

scalyvladimir commented 1 year ago

Could you pleasy clarify, what authors mean by saying Round of SST? Does it mean simply another training cycle or ?