kornia/sosnet

Pretrained weights for SOSNet, used by kornia.feature.SOSNet.

SOSNet is a 128-dimensional patch descriptor that adds a second-order similarity regularisation term to the HardNet loss, enforcing that the structure of the descriptor space reflects patch similarity at multiple scales. CVPR 2019.

Original repo: yuruntian/SOSNet

Weights

File Training data
sosnet_32x32_liberty.pth Liberty
sosnet_32x32_hpatches_a.pth HPatches-A

Citation

@InProceedings{sosnet2019,
    author    = {Tian, Yurong and Yu, Xin and Fan, Bin and Wu, Fuchao
                 and Heijnen, Huub and Balntas, Vassileios},
    title     = {{SOSNet}: Second Order Similarity Regularization for
                 Local Descriptor Learning},
    booktitle = {CVPR},
    year      = {2019}
}
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support