Research map / Video / Video segmentation and tracking
Snakes and Ladders: Two Steps Up for VideoMamba
Hui Lu, Albert A. Salah, Ronald Poppe
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 elemental residual connections to the SSM (ablation) | VideoMambaPro-M · Kinetics-400 | top-1 accuracy: 83.6 -> 84.0 | Improved |
| add masked backward computation (zeroed diagonal) to bidirectional SSM (ablation) | VideoMambaPro-M · Kinetics-400 | top-1 accuracy: 83.0 -> 84.0 | Improved |
| apply both masking and residual SSM to VideoMamba-M | VideoMamba-M · Kinetics-400 | top-1 accuracy: 82.4 -> 84.0 | Improved |
| apply both masking and residual SSM to VideoMamba | VideoMamba-M · SSv2 | top-1 accuracy: 68.3 -> 69.4 | Improved |
| apply both fixes to VideoMamba-M | VideoMamba-M · UCF-101 | accuracy: 88.2 -> 91.6 | Improved |
| apply both fixes to VideoMamba-M | VideoMamba-M · AVA V2.2 | mAP: 30.1 -> 31.9 | Improved |
| apply both fixes during ImageNet-1K pretraining | VideoMamba (Ti) · ImageNet-1K val | top-1 accuracy: 76.9 -> 78.9 | Improved |
| increase input size from 224^2 to 336^2/384^2 with 64 frames (ablation) | VideoMambaPro-M · Kinetics-400 | top-1 accuracy: 84.0 -> 85.0 | Improved |
| replace VideoMamba's learned diagonal-weight projection by masking, with fewer parameters | VideoMambaPro · Kinetics-400 | parameters: 74M -> 72M | Improved |
| scale up to much larger pretrained models (InternVideo2-1B) rather than IN-1K-only Mamba | VideoMambaPro-M · Kinetics-400 | top-1 accuracy: 84.0 vs 91.6 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- VideoMAE V2 — VideoMAE V2: Scaling Video Masked Autoencoders With Dual Masking (CVPR 2023) · better on 0 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
- Demystify Mamba in Vision: A Linear Attention Perspective (NeurIPS 2024 · similarity 0.89)
- MambaOut: Do We Really Need Mamba for Vision? (CVPR 2025 · similarity 0.88)
- Achilles' Heel of Mamba: Essential difficulties of the Mamba architecture demonstrated by synthetic data (NeurIPS 2025 · similarity 0.87)
- MoMa: Modulating Mamba for Adapting Image Foundation Models to Video Recognition (ICML 2025 · similarity 0.87)
- Mamba-Reg: Vision Mamba Also Needs Registers (CVPR 2025 · similarity 0.87)
- Vision Mamba Mender (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.
