I appreciate this challenge and providing dataset.
But as far as I checked, alignment between RAW and RGB is not so accurate on train/test data.
So I wonder same level alignment accuracy dataset will be used on validation and final test data.
(If so, I just thought blind image alignment is another big topic/issue on simple RAW2RGB mapping development.)
Best regards,