Paper Reading: Diffusion Transformer Architecture and Flow Matching

With Stable Diffusion 3 being released recently, we look at the underlying papers that make it possible. SD3 is capable of correct typography, precise prompt following and spatial reasoning, attention to fine details, and high image quality across a wide variety of styles.

We will cover:

• Recap of Vision Transformers and Diffusion Models
• Scalable Diffusion Models with Transformers (ICCV 2023 paper)
• Flow Matching for Generative Modeling (ICLR 2023 paper)
• Scaling Rectified Flow Transformers for High-Resolution Image Synthesis (Stable Diffusion 3 report)

Your host:

Hemang Chawla is an applied scientist in computer vision focusing on anti-counterfeiting at Scantrust. He has previous worked in the domains of robotics and mapping for ADAS, and has published several papers at top conferences such as ICRA, IROS, and WACV.

This event is free for all. For premium access to Taro, you may use this referral link for a 20% discount.