TorchGeo Model Integration

[isaaccorley]

Hi team! Great work with the SSL4Eco dataset, new benchmarks, and pretrained model!

I have made a PR torchgeo/torchgeo#2849 for adding the pretrained resnet50 weights to be loadable in torchgeo

Once it's merged the weights should be loadable in a few lines of code along with any additional features from the timm library:

from torchgeo.models import resnet50, ResNet50_Weights

model = resnet50(weights=ResNet50_Weights.SENTINEL2_ALL_SECO_ECO)

x = torch.randn(1, 9, 224, 224)

with torch.inference_mode():
	features = model(x)  # (1, 2048)

Let me know if anything looks incorrect in the PR!