AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

On-Policy Diffusion Reinforcement Learning Meets Off-Policy Quality Anchoring

Zunxu Liu, Zhaofan Qiu, Yazhen Xie, Yingwei Pan, Ting Yao, Tao Mei

ECCV 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 reward renormalization with off-policy samples (ablation)DiffusionNFT on OCR · OCRGPU hours: 64 -> 56 GPU hours, OCR 0.970 -> 0.969Improved
add perceptual anchored reward on top of renormalization (ablation)DiffusionCompass · OCR (out-of-domain)Aesthetic score: 4.48 -> 5.03Improved
add anchoring (off-policy) loss on top of renormalization and anchored reward (ablation)DiffusionCompass · OCROCR accuracy: 0.963 -> 0.972, Aes. 5.03 -> 4.88 is not a gainImproved
add distribution-aware filtering (ablation)DiffusionCompass · OCROCR accuracy: 0.972 -> 0.975Improved
mix on-policy and off-policy RL instead of pure on-policy (ablation)SD3.5-M · OCROCR accuracy: 0.970 -> 0.975Improved
mix on-policy and off-policy RL instead of pure off-policy (ablation)SD3.5-M · OCROCR accuracy: 0.836 -> 0.975Improved
raise off-policy loss weight lambda2 from 0.1 to 0.5 (ablation)DiffusionCompass · GenEvalGenEval overall: 0.97 -> 0.92 (OCR 0.975 -> 0.959)Hurt
set filtering strength alpha to 1000 instead of 100 (ablation)DiffusionCompass · GenEvalGenEval overall: 0.97 -> 0.94 (OCR 0.975 -> 0.961)Hurt
use SD3.5-M own initial samples as anchors instead of Qwen-Image (ablation)DiffusionCompass · OCROCR accuracy: 0.975 -> 0.973No clear effect
use Flux anchors instead of Qwen-Image (ablation)DiffusionCompass · OCRAesthetic score: 5.00 -> 4.94No clear effect
add off-policy anchoring to FlowGRPOFlowGRPO · OCROCR accuracy: 0.915 -> 0.939Improved
add off-policy anchoring to DiffusionNFTDiffusionNFT · ImageRewardImageReward: 1.64 -> 1.70 (VideoOCR 0.412 -> 0.433)Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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