Research map / Detection and segmentation / CNN features
Enabling True Global Perception in State Space Models for Visual Tasks
Jie Hui, Zhenxiang Zhang, Wenyu Mi, Jianji Wang
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 |
|---|---|---|---|
| insert GMamba modules into UNet | UNet-ConvNeXt(S) · Vaihingen | mIoU: 83.11 -> 86.00 | Improved |
| insert GMamba modules into UNet | UNet-ResNet34 · LoveDA | mIoU: 50.37 -> 52.80 | Improved |
| insert GMamba into first three stages | Faster R-CNN Swin(T) · MS-COCO detection | AP: 42.2 -> 43.7 | Improved |
| insert GMamba into first three stages | Mask R-CNN Swin(T) · MS-COCO instance segmentation | AP: 38.7 -> 39.8 | Improved |
| add SSM only (ablation) | UNet-ConvNeXt(S) · Vaihingen (UNet-ConvNeXt-Small ablation) | mIoU: 83.11 -> 84.01 | Improved |
| add frequency encoding module (FEM) (ablation) | GSSM · Vaihingen (UNet-ConvNeXt-Small ablation) | mIoU: 84.01 -> 85.30 | Improved |
| use learnable FEM and adaptive modulation instead of plain DFT (ablation) | GSSM · Vaihingen (UNet-ConvNeXt-Small ablation) | mIoU: 84.79 -> 86.00 | Improved |
| use bidirectional or four-way scans instead of single scan (ablation) | GSSM · Vaihingen (UNet-ConvNeXt-Small ablation) | mIoU: 86.00 -> 85.98 | No clear effect |
| replace GSSM with self-attention (ablation) | GMamba · Vaihingen (UNet-ConvNeXt-Small ablation) | mIoU: 86.00 -> 84.78 | Hurt |
| replace GSSM with vanilla Mamba (VSSM) (ablation) | GMamba · Vaihingen (UNet-ConvNeXt-Small ablation) | mIoU: 86.00 -> 84.01 | Hurt |
| use GMamba instead of SwinV2 module | Mask R-CNN Swin(T) · MS-COCO instance segmentation (Mask R-CNN, Swin(T)) | AP: 39.1 -> 39.8 | Improved |
Closest papers by meaning
- Spatial-Mamba: Effective Visual State Space Models via Structure-Aware State Fusion (ICLR 2025 · similarity 0.88)
- GG-SSMs: Graph-Generating State Space Models (CVPR 2025 · similarity 0.87)
- Multi-Scale VMamba: Hierarchy in Hierarchy Visual State Space Model (NeurIPS 2024 · similarity 0.87)
- GS-Bias: Global-Spatial Bias Learner for Single-Image Test-Time Adaptation of Vision-Language Models (ICML 2025 · similarity 0.87)
- CoSpace: Benchmarking Continuous Space Perception Ability for Vision-Language Models (CVPR 2025 · similarity 0.86)
- GroupMamba: Efficient Group-Based Visual State Space Model (CVPR 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.
