Research map / Video / Video-language understanding
VideoMamba: State Space Model for Efficient Video Understanding
Kunchang Li, Xinhao Li, Yi Wang, Yinan He, Yali Wang, Limin Wang, Yu Qiao
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 self-distillation from a pretrained smaller VideoMamba teacher when training the larger model (ablation) | VideoMamba-B/M · ImageNet-1K | top-1 accuracy | Improved |
| add early stopping of self-distillation teacher guidance (ablation) | VideoMamba · ImageNet-1K | top-1 accuracy | No clear effect |
| increase input resolution for ImageNet (224 to 576) | VideoMamba-M · ImageNet-1K | top-1 accuracy: 82.8 -> 84.0 | Improved |
| use attention masking instead of tube masking in masked pretraining (ablation) | VideoMamba-M · SthSth V2 | top-1 accuracy: 66.3 -> 68.5 | Improved |
| align only the last layer instead of the last 6x2 layers (ablation) | VideoMamba-M masked pretraining · SthSth V2 | top-1 accuracy: 67.7 -> 68.5 | Improved |
| use 80% mask ratio instead of 50% or 90% (ablation) | VideoMamba-M masked pretraining · SthSth V2 | top-1 accuracy: 68.1 -> 68.5 (vs 50%) | Improved |
| increase droppath from 0.1 to 0.4 in masked pretraining (ablation) | VideoMamba-M · SthSth V2 | top-1 accuracy: 68.0 -> 68.5 | Improved |
| use spatial-first scan instead of temporal-first scan (ablation) | VideoMamba-Ti · video classification | top-1 accuracy | Improved |
| increase input resolution for video tasks (ablation) | VideoMamba-Ti · K400 and SthSth V2 | top-1 accuracy | Mixed |
| increase number of input frames (ablation) | VideoMamba-Ti · K400 vs SthSth V2 | top-1 accuracy | Mixed |
| increase frames from 32 to 64 for VideoMamba-M (masked pretraining) on Breakfast (ablation) | VideoMamba-M · Breakfast | top-1 accuracy: 97.9 -> 96.9 | Hurt |
Later papers that compared with this method
7 later accepted papers report VideoMamba as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Focus-to-Perceive Representation Learning: A Cognition-Inspired Hierarchical Framework for Endoscopic Video Analysis (CVPR 2026 · better on 3 of 3 shared results)
- STDDNet: Harnessing Mamba for Video Polyp Segmentation via Spatial-aligned Temporal Modeling and Discriminative Dynamic Representation Learning (ICCV 2025 · better on 5 of 5 shared results)
- GDKVM: Echocardiography Video Segmentation via Spatiotemporal Key-Value Memory with Gated Delta Rule (ICCV 2025 · better on 3 of 3 shared results)
- LV-MAE: Learning Long Video Representations through Masked-Embedding Autoencoders (ICCV 2025 · better on 2 of 3 shared results)
- BASKET: A Large-Scale Video Dataset for Fine-Grained Skill Estimation (CVPR 2025 · better on 0 of 1 shared result)
- HierarQ: Task-Aware Hierarchical Q-Former for Enhanced Video Understanding (CVPR 2025 · better on 1 of 1 shared result)
- Smooth Regularization for Efficient Video Recognition (NeurIPS 2025 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Vision Mamba Mender (NeurIPS 2024 · similarity 0.88)
- Multi-Scale VMamba: Hierarchy in Hierarchy Visual State Space Model (NeurIPS 2024 · similarity 0.88)
- MoMa: Modulating Mamba for Adapting Image Foundation Models to Video Recognition (ICML 2025 · similarity 0.88)
- MambaOut: Do We Really Need Mamba for Vision? (CVPR 2025 · similarity 0.87)
- EfficientVMamba: Atrous Selective Scan for Light Weight Visual Mamba (AAAI 2025 · similarity 0.87)
- Mamba-Adaptor: State Space Model Adaptor for Visual Recognition (CVPR 2025 · 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.
