AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Scalable Diffusion Models with Transformers

William Peebles, Saining Xie

ICCV 2023 · Image generation › Autoregressive and latent generation · approach: High-resolution diffusion synthesis · 3,816 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use adaLN-Zero conditioning instead of in-context or cross-attention (ablation)DiT-XL/2 · ImageNet 256x256, 400K stepsFIDImproved
zero-initialize adaLN block (adaLN-Zero) instead of vanilla adaLN (ablation)DiT-XL/2 · ImageNet 256x256FIDImproved
decrease patch size (more tokens) holding model size fixed (ablation)DiT · ImageNet 256x256, 400K stepsFID-50K: DiT-S/8 153.60 -> DiT-S/2 68.40Improved
increase model size (S to XL) holding patch size fixed (ablation)DiT · ImageNet 256x256, 400K stepsFID-50K: DiT-S/8 153.60 -> DiT-L/8 118.87Improved
increase sampling steps to compensate for smaller model compute (ablation)DiT-L/2 · ImageNet 256x256, 400K stepsFID-10K: XL/2 128 steps 23.7 vs L/2 1000 steps 25.9No clear effect
add classifier-free guidance (cfg=1.50)DiT-XL/2 · ImageNet 256x256FID-50K: 9.62 -> 2.27Improved
classifier-free guidance on recallDiT-XL/2 · ImageNet 256x256Recall: 0.67 -> 0.57Hurt
replace U-Net backbone with transformer on latent patcheslatent diffusion · ImageNet 256x256FID-50K: 3.60 -> 2.27Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy