Feature request: Add HRNet-Semantic-Segmentation to the Pytorch-hub

[AlexeyAB]

Feature request: Add HRNet-Semantic-Segmentation models to the Pytorch-hub

This will be a step prior to adding HRNet to the common Pytorch models: pytorch/vision#987

The simplest way is to add hubconf.py file to GitHub repository, like: https://github.com/WongKinYiu/PyTorch_YOLOv4/blob/u5_preview/hubconf.py
So we will be able to use HRNet models by using just command like: https://pytorch.org/docs/stable/hub.html#torch.hub.help

model= torch.hub.load('WongKinYiu/PyTorch_YOLOv4:u5_preview', 'yolov4_pacsp_s', pretrained=True)
output = model(img)  # output is length 3 list

for layer in output: 
    print(layer.shape)

[PkuRainBow]

@AlexeyAB Thanks for informing us of adding our models to the Pytorch-hub and the Pytorch/vision repo.

We will follow the mentioned example and add a hubconf.py soon.

[AOS55]

This would certainly be a useful feature to compare HRNet to different architectures is there a PR related to this?

[vedal]

+1 @PkuRainBow