Design a compression strategy for local SURF descriptors using autoencoders.
Training data can be generated using the images of dataset Portello http://www.dei.unipd.it/~sim1mil/materiale/3Drecon/portelloData.zip and Castle https://github.com/openMVG/SfM_quality_evaluation/tree/master/Benchmarking_Camera_Calibration_2008/castle-P30.
Testing must be done on dataset FountainP-11 and Tiso (available at https://github.com/openMVG/SfM_quality_evaluation/tree/master/Benchmarking_Camera_Calibration_2008 and http://www.dei.unipd.it/~sim1mil/materiale/3Drecon/).
The reconstructed descriptors (only for the test set) are used to perform a SfM reconstruction using COLMAP https://colmap.github.io/, (using the two test dataset).