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

训练时,Train loss: nan Train miou: nan #41

Closed Yulian-Xie closed 3 years ago

Yulian-Xie commented 3 years ago

image

MaybeShewill-CV commented 3 years ago

@Yulian-Xie 我猜测还是数据集的问题 检查下你的数据有没有问题 标签数据读入之后数据有没有被污染

Yulian-Xie commented 3 years ago

嗯,是数据集数量太少了

MaybeShewill-CV commented 3 years ago

@Yulian-Xie ok:)