Model Training about escher HOT 2 CLOSED

can the entries in the database be weighted somehow, thus assigning them priority (or more simply, a higher probability of being picked)? If so, we could "segment" the original painting in various subpictures with high weight so that they are considered first?

from escher.

I believe a good idea is to define batches (like buckets) of images. ...... and say to the model "look for something on the batch 1" if you find something accpetable (within a threshold) stop looking, otherwise go to the images on the batch 2... see what's your error there and compare with batch 3

instead of building an entire database, to make "Tranches" or batches with pictures where it tranch will have precedence over the next one... so our images will be on the top tranch
ideally, the model will be smart enough to select the best image from the database... however.. our fine tuning should have more weight in the model.. as our painting is very very particular and has little resemblance with the other images the authors used.

from escher.