Implementation:Kornia Kornia Inlier Mask Indexing

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

Overview

Pattern for extracting verified correspondences using PyTorch boolean tensor indexing on RANSAC inlier masks.

Description

This is a user-defined pattern, not a library API. After RANSAC returns an inlier mask (boolean tensor), correspondences are filtered via standard PyTorch boolean indexing: kp1_verified = kp1[inlier_mask], kp2_verified = kp2[inlier_mask]. The pattern is simple but foundational for extracting clean correspondences for downstream geometric computation.

Usage

Apply immediately after RANSAC to separate inlier from outlier correspondences.

Code Reference

Source Location

• Repository: kornia
• File: kornia/geometry/ransac.py (forward returns inlier mask)
• Lines: L335-400

Interface Pattern

kp1_verified = kp1[inlier_mask]
kp2_verified = kp2[inlier_mask]

Import

N/A (uses PyTorch tensor indexing)

I/O Contract

Inputs

Name	Type	Required	Description
kp1	torch.Tensor	Yes	Original keypoints from image 1 of shape (N, 2)
kp2	torch.Tensor	Yes	Original keypoints from image 2 of shape (N, 2)
inlier_mask	torch.Tensor	Yes	Boolean mask of shape (N,) from RANSAC

Outputs

Name	Type	Description
kp1_verified	torch.Tensor (M, 2)	Verified keypoints from image 1, where M = sum(inlier_mask)
kp2_verified	torch.Tensor (M, 2)	Verified keypoints from image 2, where M = sum(inlier_mask)

Usage Examples

Basic Inlier Extraction

import torch
from kornia.geometry import RANSAC

ransac = RANSAC(model_type="homography", inl_th=2.0)
kp1 = torch.rand(200, 2) * 256
kp2 = torch.rand(200, 2) * 256

homography, inlier_mask = ransac(kp1, kp2)

# Extract inlier correspondences
kp1_verified = kp1[inlier_mask]
kp2_verified = kp2[inlier_mask]
print(f"Verified matches: {kp1_verified.shape[0]}")

Computing Inlier Ratio

import torch
from kornia.geometry import RANSAC

ransac = RANSAC(model_type="fundamental", inl_th=1.0)
kp1 = torch.rand(300, 2) * 480
kp2 = torch.rand(300, 2) * 480

model, inlier_mask = ransac(kp1, kp2)

inlier_ratio = inlier_mask.sum().item() / len(inlier_mask)
print(f"Inlier ratio: {inlier_ratio:.2%}")

if inlier_ratio > 0.3:
    kp1_clean = kp1[inlier_mask]
    kp2_clean = kp2[inlier_mask]

Related Pages

Implements Principle

• Principle:Kornia_Kornia_Inlier_Extraction
• Environment:Kornia_Kornia_PyTorch_Python_Environment