Research map / Video / Action recognition
MS-Temba: Multi-Scale Temporal Mamba for Understanding Long Untrimmed Videos
Arkaprava Sinha, Monish Soundar Raj, Pu Wang, Ahmed Helmy, Hieu Le, Srijan Das
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 CLIP instead of I3D visual backbone | MS-Temba · TSU | mAP: 36.1 -> 44.0 | Improved |
| add Mamba temporal encoder (ablation) | CLIP baseline · TSU | mAP: 24.7 -> 40.2 | Improved |
| add dilated SSM scanning (ablation) | MS-Temba · TSU | mAP: 40.2 -> 42.5 | Improved |
| add dilated SSM scanning (ablation) | MS-Temba · Charades | mAP: 32.4 -> 32.5 | No clear effect |
| add MS-Fuser (ablation) | MS-Temba · TSU | mAP: 42.5 -> 44.0 | Improved |
| add consistency loss Lcons alone (ablation) | MS-Temba · TSU | mAP: 42.1 -> 42.1 | No clear effect |
| add auxiliary loss Laux (ablation) | MS-Temba · TSU | mAP: 42.1 -> 43.0 | Improved |
| add auxiliary loss Laux (ablation) | MS-Temba · Charades | mAP: 33.6 -> 33.1 | Hurt |
| add both Lcons and Laux (ablation) | MS-Temba · TSU | mAP: 42.1 -> 44.0 | Improved |
| consistency loss instead of correlation loss for alignment (ablation) | MS-Temba · TSU | mAP: 43.5 -> 44.0 | Improved |
| sum + projection + SSM fusion instead of concat + projection (ablation) | MS-Fuser · TSU | mAP: 42.2 -> 44.0 | Improved |
| add a fourth Temba block (ablation) | MS-Temba · TSU and Charades | mAP | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MS-TCT — MS-TCT: Multi-Scale Temporal ConvTransformer for Action Detection (CVPR 2022) · better on 5 of 6 shared results
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
- Distillation-free Scaling of Large State-Space Models for Images and Videos (IJCV 2026 · similarity 0.88)
- Video Mamba Suite: State Space Model as a Versatile Alternative for Video Understanding (IJCV 2025 · similarity 0.88)
- VideoMamba: State Space Model for Efficient Video Understanding (ECCV 2024 · similarity 0.88)
- Backtrace Mamba: Reviving Critical Temporal Contexts via Hierarchical Memory Compression for Online Action Detection (AAAI 2026 · similarity 0.88)
- TIM: A Time Interval Machine for Audio-Visual Action Recognition (CVPR 2024 · similarity 0.86)
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 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.
