Models

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.

modelcommercialkernels time (ms),
speed-up vs original
kernels peak memory (MiB, reserved),
ratio to original
aliked-flexglueyes––
aliked-lightglueyes35.1, 1.2x1324, 0.97x
dedode-v2-byes189, 1.7x1246, 0.63x
dedode-v2-gyes332, 2.0x1984, 0.8x
disk-lightglueyes52.1, 2.2x778, 0.32x
loftr-indooryes96.2, 2.5x886, 0.46x
loftr-outdooryes95.9, 2.5x886, 0.47x
loma-byes––
loma-b128yes––
loma-b64yes––
loma-gyes––
loma-lyes––
loma-ryes––
romav1yes356, 1.1x1860, 0.45x
romav2yes354, 2.1x2300, 0.23x
xfeatyes––
xfeat-lighterglueyes––

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).

modelmodescode licenceweights licencetraining datacommercialoutputs
aliked-flexglueportable, kernelsMIT AND BSD-3-ClauseMIT AND LicenseRef-DINOv3-License AND BSD-3-Clauseunstatedyeskeypoints1, keypoints2, pairs, scores
aliked-lightglueportable, kernelsApache-2.0 AND BSD-3-ClauseApache-2.0 AND BSD-3-Clausemegadepthyeskeypoints1, keypoints2, pairs, scores
dedode-v2-bportable, kernelsMITMITmegadepthyeskeypoints1, keypoints2, pairs, scores
dedode-v2-gportable, kernelsMITMITmegadepthyeskeypoints1, keypoints2, pairs, scores
disk-lightglueportable, kernelsApache-2.0 AND BSD-3-ClauseApache-2.0 AND BSD-3-Clausemegadepthyeskeypoints1, keypoints2, pairs, scores
loftr-indoorportable, kernelsApache-2.0Apache-2.0megadepthyeskeypoints1, keypoints2, pairs, scores
loftr-outdoorportable, kernelsApache-2.0Apache-2.0megadepthyeskeypoints1, keypoints2, pairs, scores
loma-bportable, kernelsMIT AND Apache-2.0MIT AND Apache-2.0 (assumed from code)megadepthyeskeypoints1, keypoints2, pairs, scores
loma-b128portable, kernelsMIT AND Apache-2.0MIT AND Apache-2.0 (assumed from code)megadepthyeskeypoints1, keypoints2, pairs, scores
loma-b64portable, kernelsMIT AND Apache-2.0MIT AND Apache-2.0 (assumed from code)megadepthyeskeypoints1, keypoints2, pairs, scores
loma-gportable, kernelsMIT AND Apache-2.0MIT AND Apache-2.0 (assumed from code)megadepthyeskeypoints1, keypoints2, pairs, scores
loma-lportable, kernelsMIT AND Apache-2.0MIT AND Apache-2.0 (assumed from code)megadepthyeskeypoints1, keypoints2, pairs, scores
loma-rportable, kernelsMIT AND Apache-2.0MIT AND Apache-2.0 (assumed from code)megadepthyeskeypoints1, keypoints2, pairs, scores
romav1portable, kernelsMITMIT AND Apache-2.0megadepthyeswarp, certainty
romav2portable, kernelsMIT AND LicenseRef-DINOv3-LicenseMIT AND LicenseRef-DINOv3-Licenseunstatedyeswarp, certainty
xfeatportable, kernelsApache-2.0Apache-2.0megadepth, cocoyeskeypoints1, keypoints2, pairs, scores
xfeat-lighterglueportable, kernelsApache-2.0Apache-2.0megadepth, cocoyeskeypoints1, keypoints2, pairs, scores

Training datasets

iddatasetlicencecommercial use
cocoCOCO 2017 (train, instance annotations)CC-BY-4.0unclear
megadepthMegaDepth (Internet photo collections with SfM / MVS depth)CC-BY-4.0unclear