search

lhoyer / MIC

[CVPR23] Official Implementation of MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation
269 stars 40 forks source link

We have a question for your amazing work. #65

Open juajua56 opened 1 year ago

juajua56 commented 1 year ago

We greatly appreciate your amazing work; we are utilizing your model for our task, however, we are experiencing a lot of noise in the road area while using the GTA to Cityscapes setting as is.

Could you please tell me how to solve the problem or what hyperparameters I need to adjust?

The same phenomenon appears even after 40000 iterations.

However, if you use the MinEnt UDA method, the above phenomenon does not occur.

Please help me.

lhoyer commented 1 year ago

Thank you for your interest in our work!

It is hard to guess what is going wrong without knowing details about the used source and target datasets. I would recommend to start simply with DAFormer (without RCS and without FD) and to succesively activate the missing components again (RCS, FD, HRDA, and MIC) to pinpoint which UDA component might cause problems in your specific setup.