Image Matching Models (IMM)

A unified API for quickly and easily trying 50+ (and growing!) image matching models.

Jump to: Install | Use | Models | Add a Model / Contributing | Acknowledgements | Cite

Matching Examples

Compare matching models across various scenes. For example, we show SIFT-LightGlue and LoFTR matches on pairs:

(1) outdoor, (2) indoor, (3) satellite remote sensing, (4) paintings, (5) a false positive, and (6) spherical.

SIFT-LightGlue

LoFTR

Extraction Examples

You can also extract keypoints and associated descriptors.

SIFT and DeDoDe

Install

IMM can be installed directly from PyPi using pip or uv (faster)

pip install image-matching-models
# or
uv pip install image-matching-models

or, for development, clone this git repo and install with:

# Clone recursively
git clone --recursive https://github.com/gmberton/image-matching-models
cd image-matching-models

# install packages
pip install .
# or, if you want an editable install for dev work
pip install -e . 

Some models require additional optional dependencies which are not included in the default list, like torch-geometric (required by SphereGlue) and tensorflow (required by OmniGlue). To install these, use

pip install .[all]
# or 
uv pip install ".[all]"

Use

You can use any of the over 50 matchers simply like this. All model weights are automatically downloaded by the IMM.

Python API

from imm import get_matcher
from imm.viz import plot_matches, plot_kpts

# Choose any of the 50+ matchers listed below
matcher = get_matcher("superpoint-lightglue", device="cuda")
img_size = 512  # optional

img0 = matcher.load_image("assets/example_pairs/outdoor/montmartre_close.jpg", resize=img_size)
img1 = matcher.load_image("assets/example_pairs/outdoor/montmartre_far.jpg", resize=img_size)

result = matcher(img0, img1)
# result.keys() = ["num_inliers", "H", "all_kpts0", "all_kpts1", "all_desc0", "all_desc1", "matched_kpts0", "matched_kpts1", "inlier_kpts0", "inlier_kpts1"]

# This will plot visualizations for matches as shown in the figures above
plot_matches(img0, img1, result, save_path="plot_matches.png")

# Or you can extract and visualize keypoints as easily as
result = matcher.extract(img0)
# result.keys() = ["all_kpts0", "all_desc0"]
plot_kpts(img0, result, save_path="plot_kpts.png")

Command Line Interface / Standalone Scripts

You can also run matching or extraction as standalone scripts, to get the same results as above.

Matching:

# if you cloned this repo, imm_match.py is available, else see CLI below
python imm_match.py --matcher superpoint-lightglue --out-dir outputs/superpoint-lightglue --input assets/example_pairs/outdoor/montmartre_close.jpg assets/example_pairs/outdoor/montmartre_far.jpg
# or
uv run imm_match.py --matcher superpoint-lightglue --out-dir outputs/superpoint-lightglue --input assets/example_pairs/outdoor/montmartre_close.jpg assets/example_pairs/outdoor/montmartre_far.jpg

From any location where an python enviroment with IMM installed is active, you can also run

# for PyPi install, use CLI entry point
imm-match --matcher superpoint-lightglue --out-dir outputs/superpoint-lightglue --input path/to/img0 --input path/to/img2

Keypoints extraction:

# if you cloned this repo, imm_extract.py is available, else see CLI below
python imm_extract.py --matcher superpoint-lightglue --out-dir outputs/superpoint-lightglue --input assets/example_pairs/outdoor/montmartre_close.jpg
# or
uv run imm_extract.py --matcher superpoint-lightglue --out-dir outputs/superpoint-lightglue --input assets/example_pairs/outdoor/montmartre_close.jpg

From any location where an python enviroment with IMM installed is active, you can also run

# for PyPi install, use CLI entry point
imm-extract --matcher superpoint-lightglue --out-dir outputs/superpoint-lightglue --input path/to/img0

These scripts can take as input images, folders with multiple images (or multiple pairs of images), or files with pairs of images paths. To see all possible parameters run

python imm_match.py -h
# or
python imm_extract.py -h

Available Models

We support the following methods:

Dense: roma, tiny-roma, duster, master, minima-roma, ufm

Semi-dense: loftr, eloftr, se2loftr, xoftr, minima-loftr, aspanformer, matchformer, xfeat-star, xfeat-star-steerers[-perm/-learned], edm, rdd-star, topicfm[-plus]

Sparse: [sift, superpoint, disk, aliked, dedode, doghardnet, gim, xfeat]-lightglue, dedode, steerers, affine-steerers, xfeat-steerers[-perm/learned], dedode-kornia, [sift, orb, doghardnet]-nn, patch2pix, superglue, r2d2, d2net, gim-dkm, xfeat, omniglue, [dedode, xfeat, aliked]-subpx, [sift, superpoint]-sphereglue, minima-superpoint-lightglue, liftfeat, rdd-[sparse,lightglue, aliked], ripe, lisrd

See Model Details to see runtimes, supported devices, and source of each model.

Adding a new method

See CONTRIBUTING.md for details. We follow the 1st principle of PyTorch: Usability over Performance

Acknowledgements

Special thanks to the authors of all models included in this repo (links in Model Details), and to authors of other libraries we wrap like the Image Matching Toolbox and Kornia.

Cite

This repo was created as part of the EarthMatch paper. Please cite EarthMatch if this repo is helpful to you!

@InProceedings{Berton_2024_EarthMatch,
    author    = {Berton, Gabriele and Goletto, Gabriele and Trivigno, Gabriele and Stoken, Alex and Caputo, Barbara and Masone, Carlo},
    title     = {EarthMatch: Iterative Coregistration for Fine-grained Localization of Astronaut Photography},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2024},
}