eliahuhorwitz/Conffusion

Official Implementation for the "Conffusion: Confidence Intervals for Diffusion Models" paper.

24
/ 100
Experimental

This tool helps researchers and practitioners using image-to-image generative AI models, like those for super-resolution or inpainting, to get reliable results. It takes a corrupted or incomplete image and, instead of just generating a new version, provides a range of possible pixel values for the reconstructed image. This means you get a visual output along with a statistical guarantee that the true pixel value falls within the provided upper and lower bounds.

144 stars. No commits in the last 6 months.

Use this if you need to use diffusion models for image reconstruction tasks and require statistical certainty about the generated pixel values, especially in high-stakes applications.

Not ideal if your primary goal is just to quickly generate high-quality images without needing quantified uncertainty or if you are not working with diffusion models for image-to-image tasks.

image-reconstruction computer-vision-research image-quality-assurance generative-ai-reliability scientific-imaging
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 10 / 25
Maturity 8 / 25
Community 6 / 25

How are scores calculated?

Stars

144

Forks

4

Language

Python

License

Last pushed

Nov 27, 2022

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/eliahuhorwitz/Conffusion"

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