DWD Paper Drops Decoder, Diffusion Model Predicts Pixels Directly
A new paper from Gordon Wetzstein's group proposes DWD, which adds a linear head to a DiT to directly output pixels without a separate decoder, achieving a state-of-the-art pixel FID of 1.43.
2026-11-19 ~ 2026-11-19 · 2 related posts
- DWD paper: drop the decoder — linear pixel head sets pixel-SOTA FID 1.43 on ImageNet — GordonWetzstein · 2026-11-19
- DWD paper removes the decoder from diffusion models, cuts cost and hits new SOTA — Crazy-Repeat-2006 · 2026-11-19