AI Paper Atlas

Research map / Benchmarks / ReasonSeg

ReasonSeg leaderboard

39 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on ReasonSeg. Best gIoU: 73.2 (VisHarness, NeurIPS 2026).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

ReasonSeg · gIoU

24 papers, each with its best reported number for its own method; higher is better.

#gIoU ↑MethodPaperTopic
173.2VisHarness
Qwen3-VL-8B
Train the Agent, Not the Expert: Learning to Harness Heterogeneous Experts for Multi-Turn Visual Reasoning
NeurIPS 2026 · Table 2
Multimodal LLMs
272.7StAR + MV
Qwen3-VL 32B
StAR: Segment Anything Reasoner
ECCV 2026 · Table 1
Multimodal LLMs
370.1Ours (EVOG)
Qwen2.5-VL-7B
Beyond Token Representations: Explicit Visual Object Grounding for Video Reasoning Segmentation
NeurIPS 2026 · Table 3
Video segmentation and tracking
469.8VideoSeg-R1
Qwen2.5-VL-7B
VideoSeg-R1:Reasoning Video Object Segmentation via Reinforcement Learning
AAAI 2026 · Table 2
Video-language understanding
569.6Ours-4BReasoning-Guided Part-Level Visual Grounding via Reinforcement Learning
ECCV 2026 · Table 5
Multimodal LLMs
667.4Seg-ReSearch-8B
Qwen3-VL-8B
Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search
ICML 2026 · Table 2
Multimodal LLMs
767UFO-InternVL2.5-8B*
InternVL2.5-8B
UFO: A Unified Approach to Fine-grained Visual Perception via Open-ended Language Interface
NeurIPS 2025 · Table 3
Multimodal LLMs
866.73Ours (multi-object)
Qwen2.5-VL-7B
From Failure to Feedback: Group Revision Unlocks Hard Cases in Object-Level Grounding
CVPR 2026 · Table 1
Multimodal LLMs
966.5RSAgent
Qwen2.5-VL-7B
RSAgent: Learning to Reason and Act via Multi-Turn Tool Invocations for Text-Guided Segmentation
ICML 2026 · Table 2
Multimodal LLMs
1066.1DR2Seg*
Qwen2.5VL-7B
DR$^2$Seg: Decomposed Two-Stage Rollouts for Efficient Reasoning Segmentation in Multimodal Large Language Models
ICML 2026 · Table 1
Multimodal LLMs
14 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ReasonSeg · gIoU (val)

21 papers, each with its best reported number for its own method; higher is better.

#gIoU ↑MethodPaperTopic
174.6RESAnything
Pixtral 12B
RESAnything: Attribute Prompting for Arbitrary Referring Segmentation
NeurIPS 2025 · Table: Quantitative results on
Open-set and few-shot detection
273.3Seg-ReSearch-8B
Qwen3-VL-8B
Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search
ICML 2026 · Table 2
Multimodal LLMs
369RSAgent
Qwen2.5-VL-7B
RSAgent: Learning to Reason and Act via Multi-Turn Tool Invocations for Text-Guided Segmentation
ICML 2026 · Table 2
Multimodal LLMs
468.7RGA3-7B-SAM2
Qwen2.5-VL-7B
Object-centric Video Question Answering with Visual Grounding and Referring
ICCV 2025 · Table 7
Video-language understanding
568.5DR2Seg*
Qwen2.5VL-7B
DR$^2$Seg: Decomposed Two-Stage Rollouts for Efficient Reasoning Segmentation in Multimodal Large Language Models
ICML 2026 · Table 1
Multimodal LLMs
668.2SAM-Veteran
Qwen2.5-VL-7B
SAM-Veteran: An MLLM-Based Human-like SAM Agent for Reasoning Segmentation
ICLR 2026 · Table 3
Multimodal LLMs
768.1Worker-Supervisor (ours)Guideline-Consistent Segmentation via Multi-Agent Refinement
AAAI 2026 · Table 2
Open-set and few-shot detection
868.1CoReS (ft)
LLaVA-v1.5-13B
CoReS: Orchestrating the Dance of Reasoning and Segmentation
ECCV 2024 · Table 1
Multimodal LLMs
967.9UGround-13B
LLaVA1.5-13B
UGround: Towards Unified Visual Grounding with Unrolled Transformers
ICML 2026 · Table 2
Multimodal LLMs
1067.48Ours (multi-object)
Qwen2.5-VL-7B
From Failure to Feedback: Group Revision Unlocks Hard Cases in Object-Level Grounding
CVPR 2026 · Table 1
Multimodal LLMs
11 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ReasonSeg · cIoU

13 papers, each with its best reported number for its own method; higher is better.

3 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ReasonSeg · cIoU (val)

11 papers, each with its best reported number for its own method; higher is better.

1 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.