Research map / Vision-language and multimodal / Human and embodied modeling
SAM4MLLM: Enhance Multi-Modal Large Language Model for Referring Expression Segmentation
Yi-Chia Chen, Wei-Hua Li, Cheng Sun, Yu-Chiang Frank Wang, Chu-Song Chen
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 |
|---|---|---|---|
| use PQPP (proactive point query) instead of PPG (direct point generation) (ablation) | SAM4MLLM Qwen-VL-7B · refCOCOg val(U) | cIoU: 74.2 -> 74.5 | Improved |
| use PQPP instead of PPG (ablation) | SAM4MLLM Qwen-VL-7B · refCOCO val | cIoU: 76.2 -> 77.1 | Improved |
| replace Qwen-VL backbone with LLaVA1.6 for PQPP (ablation) | SAM4MLLM-PQPP · refCOCO val | cIoU: 77.1 -> 79.6 | Improved |
| set point confidence threshold to 0.9 versus 0.6 (ablation) | SAM4MLLM-PQPP · RefCOCOg val | cIoU: 69.5 -> 74.5 | Improved |
| raise threshold from 0.9 to 0.95 (ablation) | SAM4MLLM-PQPP · RefCOCOg val | cIoU: 74.5 -> 73.8 | Hurt |
| use 5x5 grid instead of random sampling of 25 points (ablation) | SAM4MLLM-PQPP · RefCOCOg val | cIoU: 73.7 -> 74.5 | Improved |
| use 6x6 grid instead of 5x5 grid (ablation) | SAM4MLLM-PQPP · RefCOCOg val | cIoU: 74.5 -> 74.4 | No clear effect |
| add more mask training data (SAM4MLLM*) with LLaVA1.6 backbone | SAM4MLLM-8B-PQPP · ReasonSeg val (zero-shot) | gIoU: 46.7 -> 58.4 | Improved |
| fine-tune MLLM with SAM prompt data | Qwen-VL-Chat-7B · VQAv2 | accuracy: 78.2 -> 78.7 | Improved |
| use SAM4MLLM instead of GLaMM on simple short queries | refCOCO / refCOCO+ vs refCOCOg · refCOCO val vs refCOCOg val | cIoU: refCOCO 77.1 vs 79.5; refCOCOg 74.5 vs 74.2 | Mixed |
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.
- GLaMM-7B — GLaMM: Pixel Grounding Large Multimodal Model (CVPR 2024) · better on 4 of 4 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
- Position Is All You Need: A Free Lunch Token Compression Strategy for MLLM-based Referring Expression Segmentation (ICML 2026 · similarity 0.89)
- X-SAM: From Segment Anything to Any Segmentation (AAAI 2026 · similarity 0.89)
- GSVA: Generalized Segmentation via Multimodal Large Language Models (CVPR 2024 · similarity 0.88)
- Text4Seg: Reimagining Image Segmentation as Text Generation (ICLR 2025 · similarity 0.88)
- Unveiling Parts Beyond Objects: Towards Finer-Granularity Referring Expression Segmentation (CVPR 2024 · similarity 0.88)
- 3D-STMN: Dependency-Driven Superpoint-Text Matching Network for End-to-End 3D Referring Expression Segmentation (AAAI 2024 · 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.
