kornia.feature

Local feature detection, description and matching, from classical operators to pretrained models, all as differentiable PyTorch modules. For guidance on which model to pick, see the image matching guide. For classical SIFT, start with the explicit scale-space pyramid recipe in the SIFT pipeline guide. For supplied patches or frames, see the SIFT descriptor interfaces.

Detectors	Keypoint detectors: Harris, GFTT, Hessian, DoG responses and the KeyNet, scale-space and multi-resolution detectors.
Descriptors	Patch, dense and supplied-frame SIFT descriptors; MKD, HardNet, HardNet8, HyNet, TFeat, SOSNet, and the SOLD2 line descriptor.
Local features (detector and descriptor together)	Detector + descriptor pipelines: DISK, ALIKED, DeDoDe, XFeat, SIFTFeatureScaleSpace, SIFTFeature, KeyNetAffNetHardNet and friends.
Matching	Nearest-neighbour, mutual, ratio-test, FGINN and AdaLAM matching; LightGlue and LoFTR.
Local affine frames (LAF)	Local affine frames (LAFs): patch extraction, normalization, orientation and affine shape estimation.
Layers and other models	Building-block layers and the DeFMO fast-moving-object model.