Research map / Detection and segmentation / CNN features
MTMamba: Enhancing Multi-Task Dense Scene Understanding by Mamba-Based Decoders
Baijiong Lin, Weisen Jiang, Pengguang Chen, Yu Zhang, Shu Liu, Yingcong Chen
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 |
|---|---|---|---|
| replace Swin Transformer decoder blocks with STM blocks (ablation) | multi-task decoder · NYUDv2 | Delta_m (%): -1.87 -> +1.84 | Improved |
| increase STM blocks from two to three (ablation) | MTMamba · NYUDv2 | Delta_m (%): +1.84 -> +1.55 | No clear effect |
| add cross-task Mamba (CTM) block (ablation) | MTMamba · NYUDv2 | Delta_m (%): +1.84 -> +2.38 | Improved |
| replace Mamba feature extractor (MFE) with window attention (ablation) | MTMamba · NYUDv2 | Delta_m (%): +2.38 -> -0.79 | Hurt |
| replace linear gate with attention-based gate (ablation) | MTMamba · NYUDv2 | Delta_m (%): +2.38 -> +2.08 | No clear effect |
| use adaptive gate in CTM instead of fixed g^t=0 or g^t=1 (ablation) | MTMamba · NYUDv2 | Delta_m (%): +1.77 / +1.76 -> +2.38 | Improved |
| scale Swin encoder from Tiny to Large (ablation) | MTMamba · NYUDv2 | Semseg mIoU: 49.25 -> 55.82 | Improved |
| use MTMamba instead of best Transformer decoders on PASCAL-Context | multi-task decoder · PASCAL-Context | Saliency maxF: 84.81 -> 84.14 (saliency) | Mixed |
| use MTMamba instead of InvPT on PASCAL-Context | multi-task decoder · PASCAL-Context | Parsing mIoU: 67.61 -> 72.62 | Improved |
Later papers that compared with this method
1 later accepted paper reports MTMamba as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Virtual Task Prompting for Multi-Task Scene Understanding (NeurIPS 2026 · better on 1 of 2 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
- MambaOut: Do We Really Need Mamba for Vision? (CVPR 2025 · similarity 0.87)
- Demystify Mamba in Vision: A Linear Attention Perspective (NeurIPS 2024 · similarity 0.87)
- Mamba as a Bridge: Where Vision Foundation Models Meet Vision Language Models for Domain-Generalized Semantic Segmentation (CVPR 2025 · similarity 0.87)
- MobileMamba: Lightweight Multi-Receptive Visual Mamba Network (CVPR 2025 · similarity 0.87)
- Mamba-Reg: Vision Mamba Also Needs Registers (CVPR 2025 · similarity 0.86)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · 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.
