loss=NAN when training - Githubissues

Skyy93 / Sample4Geo

76 stars 10 forks source link

loss=NAN when training #12

Closed dyx30 closed 6 months ago

dyx30 commented 6 months ago

Thanks for your nice work. I'm trying to train vigor_cross but loss is nan from the beginning.I checked and found that the output features of model were NAN. How can I fix that?