RIML Lab

💠 Compositional Learning Journal Club

Join us this week for an in-depth discussion on Compositional Learning in the context of cutting-edge text-to-image generative models. We will explore recent breakthroughs and challenges, focusing on how these models handle compositional tasks and where improvements can be made.

✅ This Week's Presentation:

🔹 Title: Correcting Diffusion Generation through Resampling


🔸 Presenter: Ali Aghayari

🌀 Abstract:
This paper addresses distributional discrepancies in diffusion models, which cause missing objects in text-to-image generation and reduced image quality. Existing methods overlook this root issue, leading to suboptimal results. The authors propose a particle filtering framework that uses real images and a pre-trained object detector to measure and correct these discrepancies through resampling. Their approach improves object occurrence by 5% and FID by 1.0 on MS-COCO, outperforming previous methods in generating more accurate and higher-quality images.


📄 Papers: Correcting Diffusion Generation through Resampling


Session Details:
- 📅 Date: Tuesday
- 🕒 Time: 5:30 - 6:30 PM
- 🌐 Location: Online at vc.sharif.edu/ch/rohban

We look forward to your participation! ✌️

arXiv.org

Correcting Diffusion Generation through Resampling

Despite diffusion models' superior capabilities in modeling complex distributions, there are still non-trivial distributional discrepancies between generated and ground-truth images, which has...

www.tg-me.com/jp/RIML Lab/com.RIMLLab/157

3.1K viewsAmir Kasaei, Feb 3 at 20:04

💠 Compositional Learning Journal Club

Join us this week for an in-depth discussion on Compositional Learning in the context of cutting-edge text-to-image generative models. We will explore recent breakthroughs and challenges, focusing on how these models handle compositional tasks and where improvements can be made.

✅ This Week's Presentation:

🔹 Title: Correcting Diffusion Generation through Resampling


🔸 Presenter: Ali Aghayari

🌀 Abstract:
This paper addresses distributional discrepancies in diffusion models, which cause missing objects in text-to-image generation and reduced image quality. Existing methods overlook this root issue, leading to suboptimal results. The authors propose a particle filtering framework that uses real images and a pre-trained object detector to measure and correct these discrepancies through resampling. Their approach improves object occurrence by 5% and FID by 1.0 on MS-COCO, outperforming previous methods in generating more accurate and higher-quality images.


📄 Papers: Correcting Diffusion Generation through Resampling


Session Details:
- 📅 Date: Tuesday
- 🕒 Time: 5:30 - 6:30 PM
- 🌐 Location: Online at vc.sharif.edu/ch/rohban

We look forward to your participation! ✌️

BY RIML Lab