kornia/xfeat

Pretrained weights for XFeat (Accelerated Features for Lightweight Image Matching), used by kornia.feature.XFeat.

XFeat is a lightweight keypoint detector and descriptor optimised for real-time image matching. It detects sparse keypoints and computes 64-dim descriptors in a single forward pass with a fully-convolutional architecture. CVPR 2024.

Original repo: verlab/accelerated_features

Weights

File	Description
xfeat.safetensors	Default checkpoint, safetensors format
xfeat.pt	Default checkpoint, the upstream pickle

Provenance

xfeat.safetensors holds the tensors of the upstream xfeat.pt (sha256 0f5187fd7bedd26c7fe6acc9685444493a165a35ecc087b33c2db3627f3ea10b), unchanged: the same keys, dtypes, shapes and bits. xfeat.pt here is byte-identical to that file. The kornia maintainers converted it (converter revision 5866995a4c) and checked that a model loaded from either file gives bitwise-identical outputs. The sha256 of xfeat.safetensors is d094564230ecab7d3984c0fae055dce9e80050c0fd2fdcf96b165576af5e8571.

License

Apache-2.0, the licence of verlab/accelerated_features. See LICENSE, copied from that repository.

Citation

@inproceedings{potje2024xfeat,
    author    = {Guilherme {Potje} and Felipe {Cadar} and Andre {Araujo}
                 and Renato {Martins} and Erickson R. {Nascimento}},
    title     = {{XFeat}: Accelerated Features for Lightweight Image Matching},
    booktitle = {CVPR},
    year      = {2024}
}