This is how to import an entire dataset (images and depths) from Kinect. It copies and renames the files, and reprojects and unsamples the depth maps.
Make a dataset parameters file parameters.json for the dataset. It should indicate the location of the input images and raw depth maps (not reprojected) from the Kinect, in the kinect_raw. And it should indicate the location of the final images and reprojected depth maps, in the root group.
The Kinect reprojection must have been calibrated before. But the location of the reprojection parameters file reprojection.json into the kinect_raw group, under kinect_reprojection_parameters_filename.
If images are already at the correct location, put the same value into both image_filename_format. (Assuming there is no different numbering in kinect_raw)
To import all views, run
kinect/import_raw_data.py parameters.json mine
Inside the import_raw_data.py source code, variables can be set to indicate whether to import only images or only depths, and whether to overwrite existing output files.
The depth maps will be reprojected using the given reprojection parameters, and then upsampled with the given densification method (here mine). This algorithm is implemented in src/kinect/densify/depth_densify_mine.cc.
If the environment variable LICORNEA_PARALLEL is set to 1, it will run in parallel. It will also show estimated time remaining (But this can get wrong if many files are skipped). On Linux, due to a bug in joblib (?), it can sometimes block near the final views when running in parallel. Then, terminate it, remove incomplete output files, set it to not overwrite, and re-run it non-parallel, to complete the import.