Find corresponding points between two photos of the same scene using DISK keypoints matched by LightGlue — a neural matcher from ETH Zürich introduced at ICCV 2023. Pixlane runs the full fused network locally via WebAssembly and draws score-colored match lines across a side-by-side view. Supported image contents are not uploaded for processing.
Upload
Adjust & Preview
Pixlane is loading locally. Supported local media processing, no signup, and supported file contents are not uploaded for processing.
Find corresponding points between two photos of the same scene using DISK keypoints matched by LightGlue — a neural matcher from ETH Zürich introduced at ICCV 2023. Pixlane runs the full fused network locally via WebAssembly and draws score-colored match lines across a side-by-side view. Supported image contents are not uploaded for processing.
Supported media processing runs on your device, and file contents are not uploaded for processing. Site analytics and the assistant are separate. Core access currently requires no signup; tool-specific limits may apply.
A detector network (DISK) finds distinctive keypoints in each photo, then a matcher network (LightGlue) decides which keypoints show the same physical spot. The result is a set of point-to-point correspondences with confidence scores — the building block of panorama stitching, 3D reconstruction, and visual localization.
DISK (EPFL, NeurIPS 2020) for keypoints and LightGlue (ETH Zürich, ICCV 2023) for matching, fused into one ONNX graph and run locally through ONNX Runtime WebAssembly. Both models are Apache-2.0 licensed.
Color encodes the matcher's confidence: green lines are high-confidence correspondences, yellow are mid, red are the weakest kept after your confidence filter.
No. The network weights are downloaded once (about 25 MB, cached for repeat visits) and all inference runs inside your browser. Supported image contents are not uploaded for processing.