AI Paper Atlas

Research map / Large language models / Alignment and preferences

EditReward: A Human-Aligned Reward Model for Instruction-Guided Image Editing

Keming Wu, Sicong Jiang, Max Ku, Ping Nie, Minghao Liu, Wenhu Chen

ICLR 2026 · Large language models › Alignment and preferences · approach: Reward modeling for diverse preferences

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
train Qwen2.5-VL-7B on EditReward-Data with the reward model recipeQwen2.5-VL-7B · GenAI-Benchpairwise accuracy: 40.48 -> 63.97Improved
use pair-wise uncertainty ranking loss instead of point-wise regression (ablation)EditReward · GenAI-Benchpairwise accuracy: 49.62 -> 63.97Improved
use multiple independent reward heads instead of a shared head (ablation)EditReward · GenAI-Benchpairwise accuracy: 60.17 -> 63.97Improved
use mean aggregation instead of min or sum (ablation)EditReward · GenAI-Benchpairwise accuracy: 59.96 (min), 59.63 (sum) -> 63.97Improved
use sum aggregation rather than mean (ablation)EditReward · EditReward-Benchaccuracy: 36.78 -> 37.60Mixed
increase backbone from Qwen2.5-VL-3B to 7B (ablation)EditReward · GenAI-Benchpairwise accuracy: 62.79 -> 63.97Improved
use MiMo-VL-7B backbone instead of Qwen2.5-VL-7B (ablation)EditReward · AURORA-Benchpairwise accuracy: 59.50 -> 63.62Improved
scale data from 130k to 200k and add tie disentanglement (ablation)EditReward · ImagenHubSpearman correlation: 32.45 -> 36.18Improved
scale data from 130k to 200k and add tie disentanglement (ablation)EditReward · EditReward-Benchaccuracy: 37.81 -> 36.78Hurt
fine-tune on EditReward-selected Top 20K subset instead of full noisy 46K setStep1X-Edit · GEdit-Bench-ENoverall score G_O: 6.780 -> 7.086Improved
select Top 30K or Top 10K rather than Top 20K (ablation)Step1X-Edit curation · GEdit-Bench-ENoverall score G_O: 7.086 -> 6.962 (30K), 6.938 (10K)Hurt
fine-tune Step1X-Edit on full unfiltered ShareGPT-4o-ImageStep1X-Edit · GEdit-Bench-EN-Iperceptual quality G_PQ: 6.962 -> 6.838Hurt

Later papers that compared with this method

1 later accepted paper reports EditReward as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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