AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

When Local Rules Create Global Order: Self-Organized Representation Learning for Latent Diffusion Models

Junrong Lian, Weijian Deng, Pengxu Wei, Yaqin Chen, Qixiang Ye, Liang Lin

CVPR 2026 · Image generation › Autoregressive and latent generation · approach: Latent diffusion with learned tokenizers

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
add local attraction (noise-perturbed reconstruction) to reconstruction loss (ablation)SORL · CelebA-HQ generationFID: 80.02 -> 45.54Improved
add local repulsion alone to reconstruction loss (ablation)SORL · CelebA-HQ generationFID: 80.02 -> 98.82Hurt
add local repulsion alone to reconstruction loss (ablation)SORL · CelebA-HQ generationIS: 3.13 -> 4.28Improved
combine local attraction and repulsion (ablation)SORL · CelebA-HQ generationFID: 45.54 -> 23.55Improved
increase centroid count K from 1024 to 8192 (ablation)SORL · CelebA-HQ generationFID: 28.86 -> 23.55Improved
increase centroid count K from 8192 to 16384 (ablation)SORL · CelebA-HQ generationFID: 23.55 -> 24.67No clear effect
set pull weight lambda_pull to 1e-2 instead of 1 or 1e-4 (ablation)SORL · CelebA-HQ generationFID: 45.97 (1) / 35.54 (1e-4) -> 23.55Improved
use SORL latent space versus VAE for reconstructionSORL · CelebA-HQ reconstructionPSNR: 28.40 -> 31.51Improved
use SORL latent space versus VAE for unconditional generationSORL · CelebA-HQ unconditional generationFID: 13.02 -> 9.74Improved
use SORL autoencoder for latent diffusion versus SiT-XL baselineSiT-XL/2 · ImageNet 100K stepsFID: 19.06 -> 15.12Improved
use SORL autoencoder versus iREPA on ImageNetSiT-XL/2 · ImageNet 100K stepsRecall: 0.61 -> 0.58Mixed

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