OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Supergluepretrainednetwork

Supergluepretrainednetwork is a research repository containing PyTorch code and pretrained weights for SuperGlue, a graph neural network with an optimal-matching layer that matches sparse SuperPoint keypoints and descriptors across image pairs.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Supergluepretrainednetwork — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you need a research implementation of pretrained SuperGlue to match SuperPoint features from image pairs, camera streams, directories or video, or to evaluate matches with suitable ground truth.

Who it is for

  • Researchers evaluating image feature matching
  • Developers building image-correspondence or visual-localization workflows
  • Computer-vision practitioners who need matched features for pose-estimation pipelines

Documented uses

  • Extract and visualize feature correspondences between image pairs or video frames.
  • Save keypoints, matches and confidence values for further processing.
  • Evaluate matching correctness when relative-pose ground truth and camera intrinsics are available.
  • Collate evaluation results across multiple image pairs.
Before you choose

Requirements & limits

The code and pretrained weights are provided in a public GitHub repository. No license details are supplied in the repository evidence.

  • SuperGlue operates on SuperPoint keypoints and descriptors rather than raw image pixels alone.
  • The repository includes indoor ScanNet and outdoor MegaDepth pretrained weights; the demo selects the indoor model by default.
  • Ground-truth relative poses and camera intrinsics are required for the documented match-correctness evaluation.
  • The README lists Python, PyTorch, OpenCV, Matplotlib and NumPy version requirements.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Process image pairs and save feature matches

Run the provided matcher on listed image pairs and save keypoints, matches and confidence data.

Before you start

  • Python 3.5 or later
  • PyTorch 1.1 or later
  • OpenCV 3.4 or later
  • Matplotlib 3.1 or later
  • NumPy 1.18 or later

Steps

  1. Install the documented Python dependencies: NumPy, OpenCV, PyTorch and Matplotlib.

  2. Provide the image pairs in the text-file format used by the repository.

  3. Run match_pairs.py to process the pairs and write compressed NumPy result files.

  4. Add --viz when running the script to dump visualizations of the matches.

Expected outcome

Saved result files contain keypoints, match indices and match confidence; visualization images can also be produced.

Guide limitations

  • Evaluation of match correctness requires ground-truth relative poses and camera intrinsics.
  • The repository README's default command uses its example pair set.
Documentation 1 ↗
Source-based practical guide

Run the demo on a camera or image sequence

Use the supplied demo script to view or save SuperPoint and SuperGlue feature matches from a supported input.

Before you start

  • Python 3.5 or later
  • PyTorch 1.1 or later
  • OpenCV 3.4 or later
  • Matplotlib 3.1 or later
  • NumPy 1.18 or later

Steps

  1. Install the documented Python dependencies: NumPy, OpenCV, PyTorch and Matplotlib.

  2. Run demo_superglue.py with the desired input using its --input option.

  3. For a headless image-directory run, specify --no_display and an --output_dir to save visualization images.

Expected outcome

The demo displays or saves visualized feature matches between an anchor image and incoming frames.

Guide limitations

  • The demo defaults to the indoor pretrained model.
  • The README documents CPU operation with --force_cpu; CUDA is used if a GPU is found in the default webcam example.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
magicleap/SuperGluePretrainedNetwork · Official homepage title · retained in its source language
How we review sourcesSuggest a correction