Synthetic Datasets and Evaluations for Sub-pixel Displacements Estimation from Optical Satellite Images with Deep Learning

Contains 3 synthetic datasets (UNI, DIS, UNI-5px), with three corresponding trained models, based on a CNN architecture. The three datasets contains pairs of small (either 16x16 or 32x32 pixels) windows that simulate shifts. Contains also evaluations on realistic synthetics examples of a deep learning pipeline using two of the three models presented.