Research map / Architectures and efficiency / Attention and transformers
VSSD: Vision Mamba with Non-Causal State Space Duality
Yuheng Shi, Mingjia Li, Minjing Dong, Chang Xu
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 |
|---|---|---|---|
| replace vanilla SSD with NC-SSD (ablation) | VSSD-Micro · ImageNet-1K | top-1 accuracy (%): 81.0 -> 81.6 | Improved |
| replace Bi-SSD with NC-SSD (ablation) | VSSD-Micro · ImageNet-1K | top-1 accuracy (%): 81.4 -> 81.6 | Improved |
| replace Bi-SSD with NC-SSD (ablation) | VSSD-Micro · ImageNet-1K | training throughput (imgs/sec): 399 -> 606 | Improved |
| replace vanilla SSD with NC-SSD (ablation) | VSSD-Micro · ImageNet-1K | training throughput (imgs/sec): 523 -> 606 | Improved |
| use standard attention in the last stage (ablation) | VSSD-Micro · ImageNet-1K | top-1 accuracy (%): 81.6 -> 81.8 | Improved |
| use overlapped convolutional downsampler instead of patchified (ablation) | VSSD-Micro · ImageNet-1K | top-1 accuracy (%): 81.8 -> 82.5 | Improved |
| add weight vector m to NC-SSD (ablation) | VSSD-Tiny · ImageNet-1K (100 epochs) | top-1 accuracy (%): 32.6 -> 81.8 | Improved |
| add weight vector m to NC-SSD (ablation) | VSSD-Small · ImageNet-1K (100 epochs) | training stability: training crashed in the very first epoch without m | Improved |
| use VSSD-T instead of Swin-T | Mask R-CNN 1x · COCO | box AP: 42.7 -> 46.9 | Improved |
| use VSSD-T instead of VMamba-T | UperNet · ADE20K single-scale | mIoU: 47.3 -> 47.9 | Improved |
| use VSSD-T instead of VMambaV9-T | ImageNet classifier · ImageNet-1K | top-1 accuracy (%): 82.5 -> 83.7 | Improved |
| train with MESA | VSSD-T · ImageNet-1K | top-1 accuracy (%): 83.7 -> 84.1 | Improved |
Closest papers by meaning
- Multi-Scale VMamba: Hierarchy in Hierarchy Visual State Space Model (NeurIPS 2024 · similarity 0.91)
- Spatial-Mamba: Effective Visual State Space Models via Structure-Aware State Fusion (ICLR 2025 · similarity 0.89)
- VMamba: Visual State Space Model (NeurIPS 2024 · similarity 0.89)
- Vision Mamba Mender (NeurIPS 2024 · similarity 0.88)
- 2DMamba: Efficient State Space Model for Image Representation with Applications on Giga-Pixel Whole Slide Image Classification (CVPR 2025 · similarity 0.88)
- QuadMamba: Learning Quadtree-based Selective Scan for Visual State Space Model (NeurIPS 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.
