Running workflow from notebooks and several other improvements

[franioli]

Configuration managment

• Configuration management: created new Config class that can be called with a dictionary containing the CLI parameters to parse it and setting up all the configurations. Now you can run the processing from a notebook by creating a new instance of this Config class.
• Updated input arguments in CLI for defining the paths needed for the processing. You can either define the project directory (that must contains a folder names "images") in which the results will be stores (so images and results are all in a single directory) or use the old approach defining the path to the image directory and the path to the output directory (see python main.py -h).

Back-end

• speedup matching_lowres preselection: now the extraction and matching are carried out in two steps (without repeating extraction for every pair).
• Small updates in PairsGenerator class and its methods
• Updated the interpolation function when doing image down/upsampling before extraction: replace laplacian pyramids with cv2.area for downsampling and linear or bicubic for upsampling.
• Added function to automatically create a Metashape project from a pycolmap reconstruction and run BA (it is still in the scripts folder, it is not yet integrated in Deep-Image-Matching).
• Updated to pycolmap 0.5.0 and updated the reconstruction module to adapt to it.
• Updated LightGlue to the last version to fix code crash when no keypoints are passed to the matcher (92)
• Add visualization methods for extractors and matchers. Now the keypoints and the matched features are automatically saved in the debug folder within the output directory, if the parameter -Vis passed to the parser.
• Use threading for saving features in h5 file and figures to disk
• Code cleaning in extractors and matchers modules
• Removed heavy datasets from repository. They are temporarly uploaded here.