We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
We should consider moving towards a multi-weight support API for pre-trained model weights.
Our pre-trained weight support was modeled after how torchvision used to handle weights:
model = resnet50(pretrained=True)
However, model weights are not boolean, there are many different potential weights for:
Currently, we use the following API:
model = resnet50(sensor="sentinel2", bands="all", pretrained=True)
However, this has several drawbacks:
Torchvision recently added a multi-weight support API: https://pytorch.org/blog/introducing-torchvision-new-multi-weight-support-api/
We should consider emulating torchvision using enums for all available pre-trained model weights.
Alternative is "business as usual".
I don't think this would require us to drop support for torchvision 0.12 and older, although it may make things simpler.
The text was updated successfully, but these errors were encountered:
Successfully merging a pull request may close this issue.
Summary
We should consider moving towards a multi-weight support API for pre-trained model weights.
Rationale
Our pre-trained weight support was modeled after how torchvision used to handle weights:
However, model weights are not boolean, there are many different potential weights for:
Currently, we use the following API:
However, this has several drawbacks:
Implementation
Torchvision recently added a multi-weight support API: https://pytorch.org/blog/introducing-torchvision-new-multi-weight-support-api/
We should consider emulating torchvision using enums for all available pre-trained model weights.
Alternatives
Alternative is "business as usual".
Additional information
I don't think this would require us to drop support for torchvision 0.12 and older, although it may make things simpler.
The text was updated successfully, but these errors were encountered: