memorable generative images

[dribnet]

Performed an experiment with resmem guided imagery and wanted to share some results. Here are three pairs of images created from the same starting seeds and optimized in both directions against resmem scores.

Method	Image	Score	Image	Score
vqgan snow		0.9999		0.4123
vqgan		0.9999		0.4110
vdiff		0.9995		0.4115

Interesting to me that the memorable images have a "type" (soft bumpy plumpiness?) as do the images optimized to be forgettable. Just thought you might also find this interesting - feel free to close, etc. I'm going to continue to experiment with this as I am considering adding this as a "memorability loss" to the pixray generation pipeline.