# cvg/glue-factory

Training library for local feature detection and matching

Repository: https://github.com/cvg/glue-factory
Canonical: https://ross.abutalabs.com/products/glue-factory
Language: Python
License: Apache-2.0
License Family: permissive
Topics: computer-vision, deep-learning, iccv2023, image-matching
Last push: 2026-07-20T13:55:30+00:00

## Health v2 (maintenance only)
Score: 69/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 93, release rhythm 35, longevity 75
- inputs: {"age_days": 1063, "days_push": 44, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1140, forks 157 (observed 2026-08-28T04:03:44.312910+00:00)

## What it is
Glue Factory is a PyTorch-based library for training and evaluating deep neural networks that detect and match local visual features (points and lines) in images. It supports reproducing state-of-the-art models like LightGlue and GlueStick, training on custom datasets, and benchmarking on standard datasets such as HPatches and MegaDepth-1500.

## Use cases
- train a local feature matcher for image matching
- reproduce LightGlue or GlueStick training
- evaluate feature extractors on HPatches benchmark
- train feature matching models on my own dataset
- match keypoints and lines between images with deep learning
- benchmark SuperPoint-style feature detectors
- train a model for visual localization feature matching

## When to choose
- you need to train or fine-tune deep local feature detectors or matchers
- you want reproducible state-of-the-art point/line matching models like LightGlue
- you need standardized benchmarking (HPatches, MegaDepth-1500) for feature matching
- you work in PyTorch and want auto-downloading models and datasets

## When to avoid
- you only need off-the-shelf inference without training - use LightGlue or GlueStick repos directly
- you need non-PyTorch frameworks or non-deep-learning classical feature matching (e.g. SIFT/ORB via OpenCV)
- you require permissive licensing for bundled third-party models like original SuperPoint or SuperGlue, which carry restrictive licenses

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, computer-vision, image-processing, benchmarking
- domain: computer-vision, deep-learning, machine-learning, image-processing
- platform: python, cross-platform
- tags: feature-matching, feature-detection, lightglue, gluestick, superpoint, pytorch, image-matching, training-library, hpatches, megadepth, gpu, linux, macos

## Member repositories
- cvg/glue-factory (main) score 69

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:44.312910+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:35:29.380466+00:00, confidence not recorded.
  - readme: https://github.com/cvg/glue-factory (fetched 2026-08-28T04:03:44.312910+00:00, sha e13c521f88c0)
- Data as of 2026-08-30T08:39:29.467469+00:00.