404 error when pulling model from hub

[saarimrahman]

I'm trying to pull resnext101_32x48d with the following code:

model = torch.hub.load('facebookresearch/WSL-Images', 'resnext101_32x48d_wsl')

but am receiving the following error

Downloading: "https://github.com/facebookresearch/WSL-Images/archive/master.zip" to /tmp/cache/torch/hub/master.zip
...
    raise HTTPError(req.full_url, code, msg, hdrs, fp)
urllib.error.HTTPError: HTTP Error 404: Not Found

[Gao-JT]

I have the same problem. Hope you can help to solve it. Looking forward to your reply！

[Gao-JT]

I have solved this problem. @saarimrahman The solution is:
a. manually download this code
b. unzip and put it under ~/.cache/torch/hub/
c. finally rename it to facebookresearch_WSL-Images_master. Then you can run your code successfully.

[naeem7789]

@Gao-JT we are facing the same problem.
Can you elaborate where we have to create that ~/.cache/torch/hub/?

[naeem7789]

We are running the code on kaggle.com notebook

[saarimrahman]

I ended up resolving this by downloading the checkpoint for the model from this repo, and copying over the model instantiation & definition from here. Then once I instantiated the model, I loaded the weights using the checkpoint I downloaded.

[Gao-JT]

@naeem7789 Hello, my code ran on the desktop, which may be a little different from your situation. When you use the code 'torch.hub.load()' to create the pre-trained model, the folder ‘ ~/.cache/torch/hub/’ will be created automatically to store the downloaded files and the pre-trained model checkpoint.

If your code is run on kaggle.com notebook, I think you can do the following:
a. Copy the code into your own code
b. Use the function resnext101_32x48d_wsl() copied into your code to replace the code ‘model = torch.hub.load('facebookresearch/WSL-Images', 'resnext101_32x48d_wsl')’ to load the pre-trained model.
If you still have questions, you can continue to contact me. Good luck ！

[naeem7789]

@Gao-JT , Thank you Sir , its working