metadata
license: mit
Backbones
Backbone networks exported as PyTorch programs.
All models were pretrained on ImageNet and subsequently finetuned on a specific dataset. Annotation data was converted for multi-label classification where applicable.
Usage
Simply load the model in PyTorch and run inference to get a mapping of features.
import torch
import torch.export
image = torch.randn((1, 3, 256, 256), dtype=torch.float32, requires_grad=False)
model = torch.export.load("resnet/resnet50-imagenet.pt2")
model.eval()
feats = model(image)
assert isinstance(feats, dict), type(feats)
assert feats["ext1"]
assert feats["ext2"]
assert feats["ext3"]
assert feats["ext4"]