NN predicting spatial coregistration shift of remote sensing imagery.
Predicts and applies X and Y affine transforms based on a 128x128 center crop of the input images, which have been encoded separately. Encodings are calculated based on all bands and represented as a single channel encoding.
Adapted from HighResNet - ShiftNet
10k 300x300px SPOT6 images. Download here (redacted), unpack in "data" folder in repo root. Contaings 'train' and 'val' data, out of which only train is used for now.
