Fisher Yu

@HXKwindwizard If loss is 2, it is a bad sign, saying that the model is not working properly. Probably your data is too different from what the model was trained...

ND dilated convolution is already in Caffe: https://github.com/BVLC/caffe/pull/3452

Thanks for reporting the issue. I have updated the downloading server url. drn_d_* models are available now.

The command is almost the same with Cityscapes training. Because images in BDD100K are 720p, I use 640 for crop size. So the command is like ``` python3 segment.py train...

The quoted code snippet doesn't show stride=2. For some part of the network, the downsampling layers are only for subsampled residual connection. The full information is used in the 3x3...

The code is for bilinear upsampling. In the early versions of pytorch, upsampling is aligned to the corners, which will actually cause pixel misalignment. Therefore, I made my own version...

Is this still a problem? Can you print the key on that line?

Large-scale Scene UNderstanding

Which dataset did you try to download? Which link did you use?

This seems to be `webp` support problem. `webp` needs to be installed before `Pillow` is installed. For example, if you are on Mac, you can do `brew install webp` before...