MCG-NKU/E2FGVI

about the loss

Davidcoach opened this issue · 2 comments

Hello, when I try to train your e2fgvi_hq model on the youtube-vos datasets, after 400k iterations the model will always collapse, and the flow loss will became nan. Have you ever face this problem? Looking forward to your answer!

Hello, you could refer to this issue #7.

same question,the training is not stable.