AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Optimizing Visual Generative Models via Distribution-wise Rewards

Ruihang Li, Mengde Xu, Shuyang Gu, Leigang Qu, Fuli Feng, Han Hu, Wenjie Wang

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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 distribution-wise reward RL (rejection sampling) to SiT-XL/2SiT-XL/2 · ImageNet 256FID-50K: 8.30 -> 6.98Improved
add distribution-wise reward RL to SiT-XL/2SiT-XL/2 · ImageNet 256FID-50K: 8.30 -> 5.77Improved
add distribution-wise reward RL to SiT-XL/2SiT-XL/2 · ImageNet 256FD_DINOv2: 230.39 -> 164.88Improved
use distribution-wise reward instead of sample-wise ImageRewardSiT-XL/2 RL · ImageNet 256FID: 34.26 -> 5.77Improved
add RL-EMA merging-coefficient optimization to EDM2-XSEDM2-XS · ImageNet 512FID: 3.74 -> 3.52Improved
add RL-EMA merging-coefficient optimization to EDM2-SEDM2-S · ImageNet 512FID: 2.57 -> 2.52Improved
continue RL after rejection-sampling fine-tuning (RS-then-RL) (ablation)SiT-XL/2 · ImageNet 256FID-50KNo clear effect
use batch-level instead of group-level advantage normalization (ablation)distribution-wise RL · ImageNet 256convergence speedImproved
use 5,000-image reference set instead of 7,500 (ablation)distribution-wise RL · ImageNet 256FID-50K@250Improved
replace 50 images per rollout instead of 100 or 200 (ablation)distribution-wise RL · ImageNet 256generation qualityImproved
train on global top 25% samples instead of all samples (ablation)distribution-wise RL · ImageNet 256performanceImproved
use ODE solver at evaluation for SDE-trained model (ablation)SiT SDE-trained RL · ImageNet 256FIDHurt

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