Research map / Architectures and efficiency / Attention and transformers
Spatial-Aware Reduction Framework: Towards Efficient and Faithful Visual State Space Models
Jindi Lv, Aoyu Li, Yuhao Zhou, Zheng Zhu, Xiaofeng Wang, Qing Ye, Yueqi Duan, Wentao Feng, Jiancheng Lv
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 structured spatial reduction (row/column) instead of global ToMe merging (ablation) | ToMe · ImageNet-1K VMamba-S 29% reduction | top-1 accuracy: 17.6 -> 71.4 | Improved |
| add local window constraint to structured reduction (ablation) | STORM (ToMe) · ImageNet-1K VMamba-S 36% reduction | top-1 accuracy: 65.3 -> 73.3 | Improved |
| apply STORM to ToMe on VMamba-T | ToMe · ImageNet-1K VMamba-T | top-1 accuracy: 31.5 -> 78.8 | Improved |
| apply STORM to EViT on VMamba-T | EViT · ImageNet-1K VMamba-T | top-1 accuracy: 15.2 -> 78.5 | Improved |
| apply STORM to ToMe on VMamba-B | ToMe · ImageNet-1K VMamba-B | top-1 accuracy: 35.7 -> 82.6 | Improved |
| apply STORM to ToMe on PlainMamba-L1 | ToMe · ImageNet-1K PlainMamba-L1 | top-1 accuracy: 69.6 -> 75.4 | Improved |
| apply STORM to ToMe on PlainMamba-L3 | ToMe · ImageNet-1K PlainMamba-L3 | top-1 accuracy: 76.1 -> 80.9 | Improved |
| apply STORM to EViT for detection on VMamba-B | EViT · COCO detection VMamba-B | box AP: 3.7 -> 46.2 | Improved |
| use content-aware ToMe selection instead of random selection within STORM (ablation) | STORM · ImageNet-1K VMamba-S 29% reduction | top-1 accuracy: 74.4 -> 78.0 | Improved |
| apply STORM to ToMe with 5 epochs fine-tuning | ToMe · ImageNet-1K VMamba-S | top-1 accuracy: 48.1 -> 82.1 | Improved |
| use STORM instead of Mamba-specific HSA | VMamba-S · ImageNet-1K training-free | top-1 accuracy: 46.2 -> 82.0 | Improved |
Closest papers by meaning
- Spatial-Mamba: Effective Visual State Space Models via Structure-Aware State Fusion (ICLR 2025 · similarity 0.88)
- EfficientVMamba: Atrous Selective Scan for Light Weight Visual Mamba (AAAI 2025 · similarity 0.87)
- VMamba: Visual State Space Model (NeurIPS 2024 · similarity 0.87)
- Vision Mamba Mender (NeurIPS 2024 · similarity 0.86)
- Multi-Scale VMamba: Hierarchy in Hierarchy Visual State Space Model (NeurIPS 2024 · similarity 0.86)
- MambaIC: State Space Models for High-Performance Learned Image Compression (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.
