Research map / Detection and segmentation / Open-set and few-shot detection
Rethinking MLLM Itself as a Segmenter with a Single Segmentation Token
Anqi Zhang, Xiaokang Ji, Guangyu Gao, Jianbo Jiao, Chi Harold Liu, Yunchao Wei
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 uncompressed encoder features F_HQ instead of compressed LLM features (ablation) | SELF1E · RefCOCO val | cIoU: 76.6 -> 78.7 | Improved |
| add Residual Features Refilling on top of F_HQ (ablation) | SELF1E · RefCOCO testB | cIoU: 75.2 -> 76.3 | Improved |
| add pixel-unshuffle with RFR (ablation) | SELF1E · RefCOCO val | cIoU: 79.0 -> 79.8 | Improved |
| add Residual Features Amplifier (shared MLP + separate LLM MLP) (ablation) | SELF1E · RefCOCO avg cIoU | cIoU: 78.9 -> 80.0 | Improved |
| use three independent MLPs in RFA instead of two (ablation) | SELF1E · RefCOCO avg cIoU | cIoU: 80.0 -> 79.7 | Hurt |
| pixel-unshuffle on residual features only (ablation) | SELF1E · RefCOCO avg cIoU | cIoU: 78.9 -> 79.3 | No clear effect |
| enable bidirectional [IMG]->[IMG] attention (ablation) | SELF1E · RefCOCO | cIoU: 78.4 -> 79.0 | No clear effect |
| make [SEG] visible to [IMG] tokens (ablation) | SELF1E · RefCOCO | cIoU: 78.4 -> 80.0 | Improved |
| full bidirectional mask versus [IMG]->[SEG] only (ablation) | SELF1E · RefCOCOg | cIoU: 77.4 -> 76.9 | No clear effect |
| scale SELF1E from 2B to 8B | SELF1E · ReasonSeg val | gIoU: 59.4 -> 65.9 | Improved |
| emphasize segmentation sampling (SEG version) | SELF1E-8B · RefCOCOg test | cIoU: 79.8 -> 82.8 | Improved |
| train without any data from target OVS datasets | SELF1E-2B · Pascal VOC-20 (OVS) | mIoU: 89.9 -> 44.8 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- HiMTok-8B — HiMTok: Learning Hierarchical Mask Tokens for Image Segmentation with Large Multimodal Model (ICCV 2025) · better on 3 of 3 shared results
- PSALM — PSALM: Pixelwise Segmentation with Large Multi-modal Model (ECCV 2024) · better on 1 of 1 shared result
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
- SegAgent: Exploring Pixel Understanding Capabilities in MLLMs by Imitating Human Annotator Trajectories (CVPR 2025 · similarity 0.89)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.87)
- ARGenSeg: Image Segmentation with Autoregressive Image Generation Model (NeurIPS 2025 · similarity 0.87)
- PixelLM: Pixel Reasoning with Large Multimodal Model (CVPR 2024 · similarity 0.87)
- Dense Connector for MLLMs (NeurIPS 2024 · similarity 0.86)
- Divide, Conquer and Combine: A Training-Free Framework for High-Resolution Image Perception in Multimodal Large Language Models (AAAI 2025 · similarity 0.86)
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.
