AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Reinforcing Vision-Language Models for Image Quality Assessment with Grounding Process Rewards

Jingwei Liu, Hongyan Li, Bo Liu, Tianlin Zhang, Yifei Qian, Congyang Zhao, Junhong Liu, Yang Cai, Ling Yang

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: VLM capability analysis · 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 grounding process rewards on top of outcome reward (ablation)Qwen3-VL-8B RL · Q-Bench descriptionscore: 7.60 -> 7.93Improved
RGPR instead of SFT for groundingQwen3-VL-8B · Distortion groundingAcc0.5: 31.2 -> 40.0Improved
RGPR instead of SFT for groundingQwen3-VL-8B · Distortion groundingmIoU: 39.6 -> 43.9Improved
RGPR instead of SFTQwen3-VL-8B · ViDA-UGC-Test* perceptionaccuracy: 74.03% -> 84.45%Improved
RGPR instead of VisualQuality-R1Qwen3-VL-8B · Quality assessment averageSRCC: 0.828 -> 0.831No clear effect
RGPR instead of SFTQwen3-VL-8B · ViDA-UGC-Test* descriptionoverall: 6.54 -> 7.10Improved
swap base model Qwen2.5-VL-7B to Qwen3-VL-8B under RGPRRGPR · Q-Bench descriptionoverall: 7.83 -> 7.93Improved
add step-wise reward normalization (ablation)RGPR · training dynamicsconvergenceImproved
use trajectory-level instead of token-level policy optimization (ablation)RGPR · LLVisionQA-devperformanceImproved
use maximum instead of average reward normalization function (ablation)RGPR · description taskperformanceHurt
raise grounding threshold tau above 0.5 (ablation)RGPR · groundingrewardHurt
lower grounding threshold tau below 0.3 (ablation)RGPR · groundingrewardHurt

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