I want to recognize pieces of chessboard diagram (not real 3d pieces but just diagram).
I split this task in some operation like rotation/cutting/segmentation.
First of all I want to detect chessboard rotation (x/y/z) before segmentation. I will have unlimited number of rotated chessboards with labels from photos done with camera. Photos can be done with tolerance of 5 degree for x, y (about) and 20 for z.
z is rotation of axis of lens to point of view and x, y is tilts of lens (like in normal camera).
As result it is need to get x, y, z in -180, 180 range (practically can be only -20, 20 since nobody do wrong photos).
For example like this (input is image and output is x, y, z (value with tolerance of 0.5 degree can be enough - so it can be class of rotation):
Some 2d rotation (z axis only)
Some 3d rotation (x, y, z - camera photo)
I assume that paper with chessboard diagram is not too much curved but can be.