Research map / Video / Action recognition
S2Gest: Split-Scan State Space Models for Dynamic Hand Gesture Recognition
KeFan Chen, Yong Gu, bo li, Longjie Huang, Jiajun Zhang
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 MS-Block spatial stem instead of 3D ResNet-18 (ablation) | S2Gest-Small · NVGesture Depth | Top-1 accuracy (%): 83.82 -> 90.66 | Improved |
| use S2-Block (Mamba) instead of Transformer encoder (ablation) | S2Gest-Small · NVGesture Depth | Top-1 accuracy (%): 86.51 -> 90.66 | Improved |
| add AFM guidance instead of no guidance (ablation) | S2Gest-Small · NVGesture Depth | Top-1 accuracy (%): 88.80 -> 90.66 | Improved |
| add naive element-wise addition of context instead of no guidance (ablation) | S2Gest-Small · NVGesture Depth | Top-1 accuracy (%): 88.80 -> 87.76 | Hurt |
| use AFM instead of concatenation of context (ablation) | S2Gest-Small · NVGesture Depth | Top-1 accuracy (%): 89.21 -> 90.66 | Improved |
| use Split-Scan instead of uni-directional scan (ablation) | S2Gest-Small · NVGesture Depth | Top-1 accuracy (%): 89.42 -> 90.66 | Improved |
| use Split-Scan instead of full-channel bi-directional scan (ablation) | S2Gest-Small · NVGesture Depth | Top-1 accuracy (%): 90.04 -> 90.66 | Improved |
| use Split-Scan instead of full bi-directional scan (parameters) (ablation) | S2Gest-Small · NVGesture Depth | Params (M): 3.26 -> 2.59 | Improved |
| use S2Gest-Nano instead of MobileNetV2 | lightweight backbone · NVGesture Depth | Top-1 accuracy (%): 75.73 -> 82.99 | Improved |
| use depth modality instead of RGB | S2Gest-Small · NVGesture | Top-1 accuracy (%): 86.10 -> 90.66 | Improved |
| shrink model from Small to Nano (smaller capacity) | S2Gest · Briareo RGB | Top-1 accuracy (%): 97.69 -> 98.15 | Improved |
| shrink model from Small to Nano | S2Gest · NVGesture Depth | Top-1 accuracy (%): 90.66 -> 82.99 | Hurt |
Closest papers by meaning
- Improving the Performance of Unimodal Dynamic Hand-Gesture Recognition With Multimodal Training (CVPR 2019 · similarity 0.88)
- Online Detection and Classification of Dynamic Hand Gestures With Recurrent 3D Convolutional Neural Network (CVPR 2016 · similarity 0.87)
- Prior-aware Dynamic Temporal Modeling Framework for Sequential 3D Hand Pose Estimation (ICCV 2025 · similarity 0.86)
- Deep Dynamic Neural Networks for Multimodal Gesture Segmentation and Recognition (TPAMI 2016 · similarity 0.85)
- SocialGesture: Delving into Multi-person Gesture Understanding (CVPR 2025 · similarity 0.84)
- MambaTalk: Efficient Holistic Gesture Synthesis with Selective State Space Models (NeurIPS 2024 · similarity 0.84)
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.
