deep-image-matching icon indicating copy to clipboard operation
deep-image-matching copied to clipboard

Multiview matching with deep-learning and hand-crafted local features for COLMAP and other SfM software. Supports high-resolution formats and images with rotations. Both CLI and GUI are supported.

Results 17 deep-image-matching issues
Sort by recently updated
recently updated
newest added

Hi, First thank you for this awesome project ! I wonder if it is possible to set initial camera params in the cameras.yaml file ? For example, if I define...

enhancement

Hi, thanks for the great support for multi focal length, it works fine. And I have some small questions about camera model part. 1. Why does model `simple-radial` set default...

enhancement

Use [*Hydra*](https://hydra.cc/docs/tutorials/structured_config/schema/) to manage confiuration and make it **completely equivalent** using yaml files, command line and GUI (in the future) for running DIM.

enhancement

Could be nice to not convert 16bit images to 8bit images, and check if DL local features works better

enhancement

Now it is possible that a feature in one image is matched with more than one feature in the other image (due to the overlap in tiling)

bug

HI! Thank you very much for a wonderfull job you did! I've tried to run a reconstruction python ./main.py --config superpoint+lightglue --images images --outs out --strategy sequential --overlap 2 --force...

All the detector-free matchers (loftr, se2-loftr, roma) work only on image pairs. Therefore, these approaches currentely return matches on each pair of images with multiplicity (track length) of 2. We...

enhancement

Merged matches from different local features are written in COLMAP databases both as raw and verified matches. If verified matches are deleted in COLMAP GUI and then run only geometric...

bug

Now you have to statically define the number of tiles and the Tiler compute the limits of the tiling grid. We should also give the possibility to set the tile...

enhancement

Now all the matcher classes return only the image coordinates of the matched keypoints. We should keep track also of the matching score/confidence/certainty (especially for dense/semi-dense matchers such as RoMa)

enhancement