哪一个损失函数最好用，损失函数类中里面又不带参数？

CoinCheung / pytorch-loss

label-smooth, amsoftmax, partial-fc, focal-loss, triplet-loss, lovasz-softmax. Maybe useful

MIT License

2.17k stars 374 forks source link

Closed jingzhengli closed 3 years ago

jingzhengli commented 3 years ago

你好，我看到am-softmax这个损失函数里面带着参数需要更新，这个参数实际上就是分类器的最后一层是吧。我来请教一下哪种损失函数比较容易上手呢？同时效果还比较好？

CoinCheung commented 3 years ago

maybe you can start from nn.crossentropyloss.