Research map / Video / Action recognition
VideoMamba: Spatio-Temporal Selective State Space Model
Jinyoung Park, Hee-Seon Kim, Kangwook Ko, Minbeom Kim, Changick Kim
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 |
|---|---|---|---|
| pretrain on ImageNet-1K instead of training from scratch (ablation) | VideoMamba · HMDB51 | Top-1 accuracy: 18.1 -> 59.3 | Improved |
| pretrain on Kinetics-400 instead of ImageNet-1K (ablation) | VideoMamba · HMDB51 | Top-1 accuracy: 59.3 -> 68.6 | Improved |
| pretrain on Kinetics-400 instead of ImageNet-1K (ablation) | VideoMamba · Something-Something V2 | Top-1 accuracy: 63.7 -> 63.9 | No clear effect |
| increase input frames from 16 to 32 (ablation) | VideoMamba · Kinetics-400 val | Top-1 accuracy: 76.1 -> 77.7 | Improved |
| increase input frames from 16 to 32 with K400 pretraining (ablation) | VideoMamba · HMDB51 | Top-1 accuracy: 68.6 -> 75.7 | Improved |
| use spatio-temporal reversal for backward SSM instead of spatial reversal (ablation) | VideoMamba · SSV2 and HMDB51 | Top-1 accuracy | Improved |
| add temporal expansion initialization of learnable positional embeddings (ablation) | VideoMamba · SSV2 and HMDB51 | Top-1 accuracy | Improved |
| add RandAugment and label smoothing (ablation) | VideoMamba · HMDB51 | Top-1 accuracy: about +1% each | Improved |
| reorder input frames progressively more severely (blockwise, pairwise, interleaved) (ablation) | VideoMamba · HMDB51 | Top-1 accuracy | Hurt |
| replace VideoSwin-T with VideoMamba on SSV2 | VideoMamba · Something-Something V2 val | Top-1 accuracy: 57.2 -> 63.7 | Improved |
| replace VideoSwin-T with VideoMamba on Kinetics-400 | VideoMamba · Kinetics-400 val | Top-1 accuracy: 78.8 -> 77.7 | Mixed |
Closest papers by meaning
- MoMa: Modulating Mamba for Adapting Image Foundation Models to Video Recognition (ICML 2025 · similarity 0.89)
- Spatial-Mamba: Effective Visual State Space Models via Structure-Aware State Fusion (ICLR 2025 · similarity 0.88)
- Mamba-Adaptor: State Space Model Adaptor for Visual Recognition (CVPR 2025 · similarity 0.87)
- Vision Mamba Mender (NeurIPS 2024 · similarity 0.87)
- 2DMamba: Efficient State Space Model for Image Representation with Applications on Giga-Pixel Whole Slide Image Classification (CVPR 2025 · similarity 0.87)
- VMamba: Visual State Space Model (NeurIPS 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.
