| --- |
| license: mit |
| tags: |
| - kornia |
| - feature-detection |
| - feature-description |
| - image-matching |
| --- |
| |
| # kornia/disk |
|
|
| Pretrained weights for **DISK** (Learning local features with policy gradient), |
| used by [`kornia.feature.DISK`](https://kornia.readthedocs.io/en/latest/feature.html). |
|
|
| DISK is trained with policy gradient and produces dense heatmaps of keypoint |
| probability and per-pixel 128-dim descriptors via a U-Net backbone. NeurIPS 2020. |
|
|
| **Original repo:** [cvlab-epfl/disk](https://github.com/cvlab-epfl/disk) |
|
|
| ## Weights |
|
|
| | File | Training supervision | |
| |------|----------------------| |
| | `depth-save.pth` | Depth | |
| | `epipolar-save.pth` | Epipolar geometry | |
|
|
| ## Citation |
|
|
| ```bibtex |
| @article{tyszkiewicz2020disk, |
| title = {{DISK}: Learning local features with policy gradient}, |
| author = {Tyszkiewicz, Micha{\l} and Fua, Pascal and Trulls, Eduard}, |
| journal = {Advances in Neural Information Processing Systems}, |
| volume = {33}, |
| pages = {14254--14265}, |
| year = {2020} |
| } |
| ``` |
|
|
