Implementation:Kornia Kornia Feature Matching Functions

Knowledge Sources	Kornia
Domains	Vision, Feature_Matching
Last Updated	2026-02-09 15:00 GMT

Overview

Concrete tools for matching feature descriptors between images provided by Kornia's matching module.

Description

Kornia provides three descriptor matching functions: match_nn (nearest neighbor), match_mnn (mutual nearest neighbor), and match_snn (second nearest neighbor ratio test). All accept descriptor tensors and return match indices and distances. An optional precomputed distance matrix dm can be passed to avoid recomputation. These functions use torch.cdist for efficient distance computation on GPU.

Usage

Import the appropriate matching function based on your precision/recall requirements. Use match_snn for standard Lowe's ratio test filtering.

Code Reference

Source Location

Repository: kornia
File: kornia/feature/matching.py
Lines: L81-187

Signatures

def match_nn(
    desc1: torch.Tensor,
    desc2: torch.Tensor,
    dm: Optional[torch.Tensor] = None
) -> Tuple[torch.Tensor, torch.Tensor]

def match_mnn(
    desc1: torch.Tensor,
    desc2: torch.Tensor,
    dm: Optional[torch.Tensor] = None
) -> Tuple[torch.Tensor, torch.Tensor]

def match_snn(
    desc1: torch.Tensor,
    desc2: torch.Tensor,
    th: float = 0.8,
    dm: Optional[torch.Tensor] = None
) -> Tuple[torch.Tensor, torch.Tensor]

Import

from kornia.feature import match_nn, match_mnn, match_snn

I/O Contract

Inputs

Name	Type	Required	Description
desc1	torch.Tensor	Yes	Descriptors from image 1 of shape (N1, D)
desc2	torch.Tensor	Yes	Descriptors from image 2 of shape (N2, D)
th	float	No	SNN ratio threshold, default 0.8 (match_snn only)
dm	Optional[torch.Tensor]	No	Precomputed distance matrix of shape (N1, N2)

Outputs

Name	Type	Description
match_indices	torch.Tensor (M, 2)	Index pairs mapping desc1 to desc2
match_distances	torch.Tensor (M,)	Distance for each matched pair

Usage Examples

NN Matching

import torch
from kornia.feature import match_nn

desc1 = torch.rand(100, 128)  # 100 descriptors, dim 128
desc2 = torch.rand(150, 128)  # 150 descriptors, dim 128

indices, distances = match_nn(desc1, desc2)
# indices.shape == (100, 2)
# distances.shape == (100,)

MNN Matching

import torch
from kornia.feature import match_mnn

desc1 = torch.rand(100, 128)
desc2 = torch.rand(150, 128)

indices, distances = match_mnn(desc1, desc2)
# indices contains only mutual nearest neighbor pairs

SNN with Custom Threshold

import torch
from kornia.feature import match_snn

desc1 = torch.rand(100, 128)
desc2 = torch.rand(150, 128)

indices, distances = match_snn(desc1, desc2, th=0.75)
# indices contains matches passing the ratio test

Related Pages

Implements Principle

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment