AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

EditScore: Unlocking Online RL for Image Editing via High-Fidelity Reward Modeling

Xin Luo, Jiahao Wang, Chenyuan Wu, Shitao Xiao, Xiyan Jiang, Defu Lian, Jiajun Zhang, Dong Liu, Zheng Liu

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models

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
fine-tune EditScore-7B on curated editing reward dataQwen2.5-VL-7B · EditReward-Benchpairwise accuracy (Overall): 0.432 -> 0.659Improved
self-ensemble 4 stochastic passes (Avg@4) (ablation)EditScore-72B · EditReward-Benchpairwise accuracy (Overall): 0.703 -> 0.763Improved
require reasoning before the score (ablation)EditScore-7B · EditReward-Benchpairwise accuracy (Overall): 0.621 -> 0.659Improved
widen score range from [0,10] to [0,25] (ablation)EditScore-7B · EditReward-Benchpairwise accuracy (Overall): 0.605 -> 0.659Improved
widen score range from [0,25] to [0,30] (ablation)EditScore-7B · EditReward-Benchpairwise accuracy (Overall): 0.659 -> 0.629Hurt
use raw Qwen2.5-VL-72B as RL reward instead of EditScore (ablation)OmniGen2 RL · GEdit-BenchOverall: 6.89 -> 6.42 (vs EditScore-Qwen3-VL-8B)Hurt
use raw Qwen2.5-VL-7B as RL reward (ablation)OmniGen2 RL · GEdit-BenchOverall: 6.28 -> 6.22No clear effect
RL with EditScore-7B (Avg@4) rewardOmniGen2 · GEdit-BenchOverall: 6.28 -> 6.68Improved
switch RL reward backbone to Qwen3-VL-8B (ablation)EditScore RL · GEdit-BenchOverall: 6.68 -> 6.89Improved
use 4-pass self-ensemble reward in RL (ablation)EditScore RL · RL learning curvesconvergence and final performanceImproved
RL with EditScore on Flux-Kontext-devFlux-Kontext-dev · GEdit-BenchOverall: 6.15 -> 6.87Improved
replace Flow-GRPO with TempFlow-GRPO (ablation)EditScore RL on OmniGen2 · GEdit-BenchOverall: 6.68 -> 7.21Improved

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