CoinCheung

Please notice that training labels of cityscapes are mapped from the label images pixels according to the specification. See this: https://github.com/CoinCheung/BiSeNet/blob/aa3876b4b1f2c430e07678f8c15b96465681fca0/lib/base_dataset.py#L44

@miscedence12 Did you check your dataset? You label range?

I will have a try. After going through the paper, there are some details that I feel not very clear, so it maybe not so easy to reproduce it. I...

The reason is likely to be that your image is an one-channel image, likely to be gray. I have added a bit fix, you can try again.

you can configure the batch size and the associated learning rate and training iters in the config file. It is easy to find :)

在这里: https://github.com/CoinCheung/pytorch-loss/blob/master/csrc/common.hpp

Sorry for being so late. I was busy with other things when this issue is created. Frankly, I do not have such a plan to add it because the implementation...

Hi, Would you roll back to cuda 10.2 and try again? I have not tried cuda11, there might be unknown problems.

I write cuda kernels because it would be a bit faster and memory efficient in some occasions. For example, when you implement label-smooth cross entropy with pytorch, you might need...

Have you ever tried to add `model.eval()`? It is some pytorch knowledge. If you run `nn.BatchNorm1d` in the training mode, you should ensure batchsize to be greater than 1. If...