ZiyiZhang27/tdpo

[ICML 2024] Code for the paper "Confronting Reward Overoptimization for Diffusion Models: A Perspective of Inductive and Primacy Biases"

26
/ 100
Experimental

This project helps AI researchers and practitioners fine-tune diffusion models to generate higher-quality images. It takes a pre-trained diffusion model and a reward function (like aesthetic score or PickScore) as input, then outputs an optimized model that produces images better aligned with human preferences, without over-optimizing for specific rewards. Researchers working with generative AI for image creation would find this useful.

No commits in the last 6 months.

Use this if you are developing or experimenting with diffusion models and need to fine-tune them more effectively to align with diverse human preferences, especially when aiming for generalization across different quality metrics.

Not ideal if you are looking for a plug-and-play solution for general image editing or generation without deep involvement in model training and research.

Generative AI Diffusion Models Image Generation AI Research Machine Learning Engineering
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 7 / 25
Maturity 16 / 25
Community 3 / 25

How are scores calculated?

Stars

38

Forks

1

Language

Python

License

MIT

Last pushed

Jul 12, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/ZiyiZhang27/tdpo"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.