search

ExplainableML / ReNO

ReNO: Enhancing One-step Text-to-Image Models through Reward-based Noise Optimization
MIT License
64 stars 3 forks source link

Aesthetic score #5

Open jsonBackup opened 1 month ago

jsonBackup commented 1 month ago

I'm curious about how the aesthetic scores of the generated images from the paper were calculated. Is it possible to share the code?

Thanks!

sgk98 commented 3 weeks ago

Hey, sorry for the delayed response. We were using the LAION Aesthetic Predictor (https://github.com/christophschuhmann/improved-aesthetic-predictor). You can find the class definition here (https://github.com/ExplainableML/ReNO/blob/main/rewards/aesthetic.py). The only difference is we return 10.0 - aesthetic_score if you want a loss to optimize. Do let us know if you need any other details, or a script to compute the aesthetic scores of all the images in a directory.