Bisenet v2 does not work when using the cityscape dataset, the result has always been NAN, and using my own dataset miou is only 0.49

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

Bisenet v2 does not work when using the cityscape dataset, the result has always been NAN, and using my own dataset miou is only 0.49 #10

Closed iloyu closed 4 years ago

iloyu commented 4 years ago

MaybeShewill-CV commented 4 years ago

@iloyu 1. Make sure you have correctly generated the tensorflow records 2. Make sure you have the same tf version as mine. 3.Use eval scripts to evaluate the miou on your test dataset:)

iloyu commented 4 years ago

I have advanced tensorflow version，is that matters？

iloyu commented 4 years ago

in the log ,it says generate ok!

MaybeShewill-CV commented 4 years ago

@iloyu The different version of tf may cause the problem. Such a problem is not supposed to occur if you have the same version of tf as mine:)