SatlasPretrain: ResNet50/152 and Swin_V2_T Weights #2038
Conversation
github-actions
bot
added
documentation
There was a problem hiding this comment.
Need to apply all the same changes I requested on #1884
Also, CI passes because we only use fake model weights, but using the real model weights result in test failures:
> pytest -m slow tests/models/test_resnet.py
...
E RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.
github-actions
bot
added
testing
adamjstewart
changed the title
| num_channels = weights.meta['in_chans'] | ||
| out_channels = model.features[0][0].out_channels | ||
| # https://github.com/allenai/satlaspretrain_models/blob/main/satlaspretrain_models/models/backbones.py#L27 | ||
| model.features[0][0] = torch.nn.Conv2d( |
There was a problem hiding this comment.
This was completely broken for MS models until now.
| @@ -272,3 +272,54 @@ def apply_transform( | |||
| out = rearrange(out, 'b t c h w -> (b t) c h w') | |||
|
|
|||
| return out | |||
|
|
|||
|
|
|||
| class _Clamp(K.IntensityAugmentationBase2D): | |||
There was a problem hiding this comment.
@johnnv1 is this something Kornia would be interested in? If so I can contribute it when I have a chance.
There was a problem hiding this comment.
I think so, not sure why it will be a augmentation. But i think it's fine to add to kornia if have uses for it. Just to make sure, @shijianjian we don't have it, right?
|
Are there any published performance metrics for all weights? Would love to add these to the weights tables. |
Looking into if we have performance metrics that could fit into the torchgeo tables. Everything else looks good to me. Thank you both so much for making this PR! |
It doesn't have to be the same benchmark datasets we used, it could be different benchmarks.
I'm planning on releasing TorchGeo 0.6.0 by the end of this month, so hopefully he can review by Monday. |
|
Going to merge so we can finalize the release soon. @favyen2 can review later if he has time. |
This PR adds SatlasPretrain ResNet50 weights as the Enums.
@piperwolters @favyen2