the loss is always nan - Githubissues

MaybeShewill-CV / bisenetv2-tensorflow

Unofficial tensorflow implementation of real-time scene image segmentation model "BiSeNet V2: Bilateral Network with Guided Aggregation for Real-time Semantic Segmentation"

https://maybeshewill-cv.github.io/bisenetv2-tensorflow/

MIT License

224 stars 59 forks source link

the loss is always nan #53

Closed huangqi111 closed 2 years ago

huangqi111 commented 3 years ago

4:36:20 Train loss: nan Train miou: nan Val loss: 15.78061 Val miou: 0.00047...

when train，the loss is always nan ，how to solve this problem

MaybeShewill-CV commented 3 years ago

@huangqi111 Most caused by label. Check label first:)