AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

Reasoning via Implicit Self-supervised Emergence for Instruction Segmentation

Qing Zhou, Lichang Yang, Yuyu Jia, Junyu Gao, Weiping Ni, Junzheng Wu, Qi Wang

AAAI 2026 · Large language models › Reasoning and chain of thought · approach: Efficient and adaptive thinking · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes RISE, a framework that learns compositional reasoning for instruction-guided segmentation via reinforcement learning with GRPO and no ground-truth masks.

Problem

Instruction-guided segmentation tasks are assumed to require complex, explicit mask supervision; the paper challenges this assumption.

Approach

Learns intricate compositional reasoning (spatial relations to world knowledge) for segmentation through self-supervised reinforcement learning with GRPO, without any ground-truth masks.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add CLIP semantic reward to format rewards (ablation)RISE (3B) · ReasonSeg testgIoU: 28.6 -> 52.9Improved
add format rewards to semantic reward (ablation)RISE (3B) · ReasonSeg testgIoU: 51.5 -> 52.9Improved
add think format reward (ablation)RISE (3B) · ReasonSeg testcIoU: 41.1 -> 45.1Improved
add segmentation format reward (ablation)RISE (3B) · ReasonSeg testcIoU: 42.5 -> 45.1Improved
add explicit chain-of-thought block (ablation)RISE (3B) · ReasonSeg testcIoU: 38.5 -> 45.1Improved
add explicit chain-of-thought block (ablation)RISE-ClsOnly (3B) · ReasonSeg testcIoU: 32.0 -> 41.7Improved
require auxiliary points in output (ablation)RISE (3B) · ReasonSeg testcIoU: 42.8 -> 45.1Improved
require auxiliary points in output (ablation)RISE (3B) · ReasonSeg testgIoU: 52.2 -> 52.9No clear effect
scale reasoning agent from 3B to 7BRISE · ReasonSeg testgIoU: 52.9 -> 58.7Improved
train on 2K ImageNet class labels instead of RefCOCOg expressionsRISE-7B · ReasonSeg testgIoU: 58.7 -> 54.5Hurt
RISE self-supervised RL on top of Qwen2.5VL-7B + SAM2Qwen2.5VL-7B + SAM2 · RefCOCO+ testAcIoU: 73.5 -> 77.7Improved
RISE-3B-ClsOnly vs base Qwen2.5VL-3BQwen2.5VL-3B + SAM2 · RefCOCO testAcIoU: 75.9 -> 76.3Improved

Surprising findings

Experimental setup

Builds on: GRPO

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