search

yu-changqian / TorchSeg

Fast, modular reference implementation and easy training of Semantic Segmentation algorithms in PyTorch.
MIT License
1.4k stars 253 forks source link

Focal loss #16

Open lingtengqiu opened 5 years ago

lingtengqiu commented 5 years ago

我感觉你的focal loss 有点和论文不一样是你自己修改的还是,这么样的呢?可以给我些解答吗

Lukzin commented 5 years ago

I feel the same way with @lingtengqiu. The SigmoidFocalLoss is not correctly implemented.

yu-changqian commented 5 years ago

There is a little problem. I will fix it and re-run the experiments.