Train w/ 64x64 crops and average at prediction time #5
Description
Training with 64x64 crops would erase the need for a 200 embedding of all proteins and would erase also noisy predictions since there would be an averaging at prediction time
{{ message }}