AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

QPrompt-R1: Real-Time Reasoning for Domain-Generalized Semantic Segmentation via Group-Relative Query Alignment

Fengyuan Lu, Zixuan Duan, Xunzhi Xiang, Zhicheng Zhang, Wenbin Li, Yang Gao, Qi Fan

ICLR 2026 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Domain-generalized segmentation

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 QPrompt learnable queries in final ViT block instead of MLP head (ablation)DINOv2-L segmentation · GTAV->Citys+BDD+Map avgmIoU: 59.9 -> 62.3Improved
add image-prototype alignment loss (ablation)QPrompt · GTAV->Citys+BDD+Map avgmIoU: 62.3 -> 62.6Improved
add group-relative alignment reward (ablation)QPrompt · GTAV->Citys+BDD+Map avgmIoU: 62.6 -> 63.7Improved
add KL divergence to EMA reference model (ablation)QPrompt-R1 · GTAV->Citys+BDD+Map avgmIoU: 63.7 -> 64.1Improved
add GRQA as plug-in trainingREIN · GTAV->Real avgmIoU: 64.3 -> 65.5Improved
add GRQA as plug-in trainingSoMA · GTAV->Real avgmIoU: 68.2 -> 68.8Improved
use GRQA reward instead of DINO-R1 reward (ablation)QPrompt · GTAV->Citys+BDD+Map avgmIoU: 62.9 -> 64.1Improved
add any reward (DINO-R1) versus none (ablation)QPrompt · GTAV->Citys+BDD+Map avgmIoU: 62.3 -> 62.9Improved
add QPrompt then GRQA on CLIP-L backbone (ablation)CLIP-L · GTAV->CitysmIoU: 50.4 -> 51.7 -> 53.2Improved
add QPrompt then GRQA on SAM-H backbone (ablation)SAM-H · GTAV->CitysmIoU: 54.7 -> 55.8 -> 57.2Improved
replace Mask2Former transformer decoder with MLP head (ablation)Mask2Former · GTAV->CitysmIoU: 63.7 -> 61.3Hurt
remove pixel decoder from Mask2Former (ablation)Mask2Former · GTAV->CitysmIoU: 63.7 -> 62.9Mixed

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