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.
Renamed core.py to matcher_base.py and updated repository structure
Added hloc code inside deep_image_matching
Recovered full matcher behaviour for matching an image pair (no return for exiting the Matcher.match() method anymore)
Added possibility to provide custom features to Lightglue matcher as parameter to the match method
Matching per tile works with all the different behaviors (TileSelection.NONE, TileSelection.GRID, TileSelection.PRESELECTION that can be setted in config.py), BUT the order of the keypoints extracted on the same image is not kept for different image pairs. Therefore, the export to COLMAP for SfM DON'T work yet.
Updated procedure for passing the configuration to the matcher: now you can pass the confiuration dictionary both to the constructor (init) or to the match() method and they will update the default configuration stored in the class.
TODO:
separating feature extraction by feature matching in such a way as the features are first extracted on ALL the images (either by tile or on full images) and then the matching routing returns only the index of the matched features.
Both the extraction and the matching must work per tiles (on the same tiles)
TODO: