superMatch 0.0.5

Local Feature Extractors and Matchers Network Library for PyTorch

dloc - the deeplearning localization toolbox

This is dloc, a modular toolbox for state-of-the-art local feature extaction and matching.

Support

Reference

• d2net: extract keypoint from 1/8 feature map
• superpoint: could extract in corner points, pretrained with magicpoint
• superglue: excellent matching algorithm but pretrained model only support superpoint, we have implementation superglue with sift/superpoint in megadepth datasets.
• disk: add reinforcement for keypoints extraction
• aslfeat: build multiscale extraction network
• cotr: build transformer network for points matching
• loftr: dense extraction and matching with end-to-end network
• r2d2: add repeatability and reliability for keypoints extraction
• contextdesc: keypoints use sift, use full image context to enhence descriptor. expensive calculation.
• NGRANSAC

If you are interested in local feature, https://drive.weixin.qq.com/s?k=AJEAIQdfAAo0ta3HBsAKMA9AatACk is useful.

Installation

pip install superMatch

dloc could run on mirrors.tencent.com/xlab/colmap:v2.0 with python3.

Debugging and Visualization

Download https://drive.weixin.qq.com/s?k=AJEAIQdfAAob00fTSEAKMA9AatACk datasets to assets.

Weights could be download from https://drive.weixin.qq.com/s?k=AJEAIQdfAAo97Nnovq.

industrial visualization and debug

## initial model
model, config = dloc.build_model('superpoint_aachen', 'superglue_outdoor', 'SuperMatch/weights', landmark=False, direct=False)
## get matches or relative pose
results = dloc.get_matches('SuperMatch/assets/2543051003_7b970fc234_o.jpg', 'SuperMatch/assets/301490830_aded42bc67_o.jpg', model, config, landmarks=None)
or
results_pose = dloc.get_pose('SuperMatch/assets/2543051003_7b970fc234_o.jpg', 'SuperMatch/assets/301490830_aded42bc67_o.jpg', model, config, mode='H')

explanation

build_model(extractor, matcher, model_path='', landmark=False, direct=False)
get_matches(name0, name1, model, config, resize=[-1], with_desc=False, landmarks=None)
get_pose(img1, img2, model, config, resize=[-1], landmarks=None, mode='H')

• extractors: 'superpoint_aachen', 'superpoint_inloc', 'd2net-ss', 'r2d2-desc','context-desc', 'landmark', 'aslfeat-desc', 'disk-desc', 'swin-disk-desc'
• matchers: 'superglue_outdoor', 'superglue_disk', 'superglue_swin_disk', 'superglue_indoor', 'NN', 'disk', 'cotr', 'loftr'