Research map / Vision-language and multimodal / Multimodal LLMs
GEAR-Seg: A Grounded Explainable Agent for Reasoning Segmentation and Data Engine
Yanan Wang, Wen li, Yibin Ying, Zhenghao Fei
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add DAM dense descriptions instead of LLaMA captions (ablation) | GEAR-Seg · ReasonSeg val | gIoU: 35.3 -> 57.4 | Improved |
| add DAM dense descriptions instead of LLaMA captions (ablation) | GEAR-Seg · LLM-Seg40k val | gIoU: 30.4 -> 52.2 | Improved |
| swap DeepSeek-R1 for Gemini-3.1-Pro as reasoning core | GEAR-Seg · ReasonSeg val | gIoU: 57.4 -> 58.2 | Improved |
| swap DeepSeek-R1 for Gemini-3.1-Pro as reasoning core | GEAR-Seg · LLM-Seg40k val | gIoU: 52.2 -> 48.5 | Hurt |
| swap DeepSeek-R1 for Qwen3-32B | GEAR-Seg · ReasonSeg val | gIoU: 57.4 -> 53.9 | Hurt |
| swap DeepSeek-R1 for local Qwen 8B | GEAR-Seg · StrawDI | mAP50:95: 0.507 -> 0.512 | No clear effect |
| swap DeepSeek-R1 for local Qwen 8B | GEAR-Seg · Mega_Peach | mAP50:95: 0.368 -> 0.345 | Hurt |
| swap DeepSeek-R1 for Sentence-Transformers | GEAR-Seg · StrawDI | mAP50:95: 0.507 -> 0.456 | Hurt |
| add explanation text loss (L_txt) to mask loss (ablation) | LISA-13B on GEAR-131K · GEAR-131K test | gIoU: 26.6 -> 37.2 | Improved |
| train on 1.2k GEAR-Seg images instead of 239 | LISA-13B distilled · ReasonSeg val | gIoU: 51.4 -> 57.2 | Improved |
| use GEAR-Seg pseudo-labels instead of human labels (239 images) | LISA-13B distilled · ReasonSeg val | gIoU: 55.6 -> 51.4 | Hurt |
| use GEAR-Seg zero-shot instead of fine-tuned LISA-13B | GEAR-Seg · ReasonSeg val | cIoU: 62.9 -> 47.5 | Hurt |
Closest papers by meaning
- DR$^2$Seg: Decomposed Two-Stage Rollouts for Efficient Reasoning Segmentation in Multimodal Large Language Models (ICML 2026 · similarity 0.90)
- SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation (CVPR 2026 · similarity 0.89)
- Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search (ICML 2026 · similarity 0.88)
- Fast Reasoning Segmentation for Images and Videos (CVPR 2026 · similarity 0.88)
- CoPRS: Learning Positional Prior from Chain-of-Thought for Reasoning Segmentation (ICLR 2026 · similarity 0.87)
- PixelCraft: A Multi-Agent system for High-Fidelity Visual Reasoning on Structured Images (ICLR 2026 · similarity 0.87)
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.
