Matching models
17 matching models. Each section says what the model is, whether it may be used commercially and how fast it is; licences and checks are under Details.
Input: two RGB images of one scene. Output: for every pixel of the first image its position in the second (x, y in pixels) and a certainty that the pixel is visible in both. Sparse matchers (LightGlue) return the keypoints of both images and the index pairs of the matched ones instead. To reconstruct camera poses from an image folder with these matchers, see structure from motion.
Measured on RTX 5060 Ti, torch 2.14.1+cu130; fastforks bench MODEL --size HxW measures yours. The summary uses one input, 768x1024: kernels time with its speed-up over the faster original run, and kernels peak reserved memory with its ratio to the smaller original one. Where measured (one variant per family), a variant's table also lists portable under torch.compile and the fp16 TensorRT engine of portable (export; memory: the engine's device memory + torch's peak). – = not measured; no kernels = the family has no kernels backend (reason under its table). Reserved memory is the allocator's cache: kernels (CUDA graphs, cuDNN workspaces) can reserve a few percent to some tens of percent more than portable at the same allocated peak.
Other devices: Apple M1 Pro 16 GB, NVIDIA L4, NVIDIA L40S, NVIDIA A10, NVIDIA A100-SXM4-40GB, NVIDIA RTX PRO 6000; across devices.
| model | commercial | kernels time (ms), speed-up vs original | kernels peak memory (MiB, reserved), ratio to original |
|---|---|---|---|
aliked-flexglue | yes | – | – |
aliked-lightglue | yes | 35.1, 1.2x | 1324, 0.97x |
dedode-v2-b | yes | 189, 1.7x | 1246, 0.63x |
dedode-v2-g | yes | 332, 2.0x | 1984, 0.8x |
disk-lightglue | yes | 52.1, 2.2x | 778, 0.32x |
loftr-indoor | yes | 96.2, 2.5x | 886, 0.46x |
loftr-outdoor | yes | 95.9, 2.5x | 886, 0.47x |
loma-b | yes | – | – |
loma-b128 | yes | – | – |
loma-b64 | yes | – | – |
loma-g | yes | – | – |
loma-l | yes | – | – |
loma-r | yes | – | – |
romav1 | yes | 356, 1.1x | 1860, 0.45x |
romav2 | yes | 354, 2.1x | 2300, 0.23x |
xfeat | yes | – | – |
xfeat-lighterglue | yes | – | – |
All models
Modes, licences, training data and outputs per model
Commercial status per variant: yes if the code and the weights licence both allow commercial use (the training data is listed but does not change it; weights without a licence of their own are assumed to carry the code licence); no if either forbids it; unclear for custom licence terms (treat as no).
| model | modes | code licence | weights licence | training data | commercial | outputs |
|---|---|---|---|---|---|---|
aliked-flexglue | portable, kernels | MIT AND BSD-3-Clause | MIT AND LicenseRef-DINOv3-License AND BSD-3-Clause | unstated | yes | keypoints1, keypoints2, pairs, scores |
aliked-lightglue | portable, kernels | Apache-2.0 AND BSD-3-Clause | Apache-2.0 AND BSD-3-Clause | megadepth | yes | keypoints1, keypoints2, pairs, scores |
dedode-v2-b | portable, kernels | MIT | MIT | megadepth | yes | keypoints1, keypoints2, pairs, scores |
dedode-v2-g | portable, kernels | MIT | MIT | megadepth | yes | keypoints1, keypoints2, pairs, scores |
disk-lightglue | portable, kernels | Apache-2.0 AND BSD-3-Clause | Apache-2.0 AND BSD-3-Clause | megadepth | yes | keypoints1, keypoints2, pairs, scores |
loftr-indoor | portable, kernels | Apache-2.0 | Apache-2.0 | megadepth | yes | keypoints1, keypoints2, pairs, scores |
loftr-outdoor | portable, kernels | Apache-2.0 | Apache-2.0 | megadepth | yes | keypoints1, keypoints2, pairs, scores |
loma-b | portable, kernels | MIT AND Apache-2.0 | MIT AND Apache-2.0 (assumed from code) | megadepth | yes | keypoints1, keypoints2, pairs, scores |
loma-b128 | portable, kernels | MIT AND Apache-2.0 | MIT AND Apache-2.0 (assumed from code) | megadepth | yes | keypoints1, keypoints2, pairs, scores |
loma-b64 | portable, kernels | MIT AND Apache-2.0 | MIT AND Apache-2.0 (assumed from code) | megadepth | yes | keypoints1, keypoints2, pairs, scores |
loma-g | portable, kernels | MIT AND Apache-2.0 | MIT AND Apache-2.0 (assumed from code) | megadepth | yes | keypoints1, keypoints2, pairs, scores |
loma-l | portable, kernels | MIT AND Apache-2.0 | MIT AND Apache-2.0 (assumed from code) | megadepth | yes | keypoints1, keypoints2, pairs, scores |
loma-r | portable, kernels | MIT AND Apache-2.0 | MIT AND Apache-2.0 (assumed from code) | megadepth | yes | keypoints1, keypoints2, pairs, scores |
romav1 | portable, kernels | MIT | MIT AND Apache-2.0 | megadepth | yes | warp, certainty |
romav2 | portable, kernels | MIT AND LicenseRef-DINOv3-License | MIT AND LicenseRef-DINOv3-License | unstated | yes | warp, certainty |
xfeat | portable, kernels | Apache-2.0 | Apache-2.0 | megadepth, coco | yes | keypoints1, keypoints2, pairs, scores |
xfeat-lighterglue | portable, kernels | Apache-2.0 | Apache-2.0 | megadepth, coco | yes | keypoints1, keypoints2, pairs, scores |
Training datasets
| id | dataset | licence | commercial use |
|---|---|---|---|
coco | COCO 2017 (train, instance annotations) | CC-BY-4.0 | unclear |
megadepth | MegaDepth (Internet photo collections with SfM / MVS depth) | CC-BY-4.0 | unclear |