Hang Zhang

Simply changing this to regular BN https://github.com/zhanghang1989/PyTorch-Encoding/blob/master/experiments/segmentation/train.py#L150

找到了当年推导 SyncBN 的手稿，要换电脑了，发到这里留个纪念


The settings are the same as PyTorch pretrained model: All pre-trained models expect input images normalized in the same way, i.e. mini-batches of 3-channel RGB images of shape (3 x...

The model is trained and tested using Gluon MXNet. The main difference is the data pipeline. They use opencv for load and preprocess images.

when loading the same image and resizing it using bilinear mode, PIL and opencv won't give the same output.

Not sure yet. If the package has better impact (more people using it), I will add support for more models.

I agree @HITjialanyu , but I mainly working on segmentation area. Will working on it if I have spare time.

This sounds easy. Will do in later version.

Will add cityscapes in GluonCV in later release.