Research map / Video / Action recognition
MMAD: Multi-label Micro-Action Detection in Videos
Kun Li, Pengyu Liu, Dan Guo, Fei Wang, Zhiliang Wu, Hehe Fan, Meng Wang
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 |
|---|---|---|---|
| replace the temporal-informative adapter with the dual-path spatial-temporal adapter (DSTA) | AdaTAD with VideoMAE-S · MMA-52 action-level | mAP Avg: 16.25 -> 18.16 | Improved |
| replace AdaTAD with DSTA | AdaTAD with VideoMAE-B · MMA-52 (action/body average) | AVG mAP: 23.08 -> 25.34 | Improved |
| replace AdaTAD with DSTA | AdaTAD with VideoMAE-S · MMA-52 (action/body average) | AVG mAP: 21.80 -> 23.43 | Improved |
| add spatial path to temporal adapter (DSTA vs TIA) | AdaTAD with VideoMAE-S · MMA-52 action-level | mAP@0.7: 10.93 -> 9.03 | Mixed |
| use DSTA instead of full fine-tuning (ablation) | VideoMAE-S snippet features · MMA-52 action-level | mAP: 15.40 -> 18.16 | Improved |
| use DSTA instead of a standard adapter (ablation) | VideoMAE-S · MMA-52 action-level | mAP: 15.87 -> 18.16 | Improved |
| adapt the backbone (full fine-tuning) rather than frozen snippet features (ablation) | VideoMAE-S · MMA-52 action-level | mAP: 10.60 -> 15.40 | Improved |
| insert adapter in all layers rather than only layers 1-4 (ablation) | DSTA VideoMAE-S · MMA-52 action-level | mAP Avg: 10.38 -> 18.16 | Improved |
| insert adapter in middle or latter layers rather than early layers (ablation) | DSTA VideoMAE-S · MMA-52 action-level | mAP Avg: 10.38 -> 13.83 | Improved |
| use larger backbone VideoMAE-B instead of VideoMAE-S | DSTA · MMA-52 (action/body average) | AVG mAP: 23.43 -> 25.34 | Improved |
| use multi-label detectors (MS-TCT, PointTAD) rather than feature-based single-label detectors | MMAD baselines · MMA-52 (action/body average) | AVG mAP: 5.64 and 8.32 vs 20.34 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- TriDet — TriDet: Temporal Action Detection With Relative Boundary Modeling (CVPR 2023) · better on 1 of 1 shared result
- MS-TCT — MS-TCT: Multi-Scale Temporal ConvTransformer for Action Detection (CVPR 2022) · better on 1 of 1 shared result
- ActionFormer — ActionFormer: Localizing Moments of Actions with Transformers (ECCV 2022) · 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
- Amplify, Aggregate, and Adjust: VideoMAE-based Holistic-Subtle Aggregation for Micro-Action Recognition (ECCV 2026 · similarity 0.91)
- MA-Bench: Towards Fine-grained Micro-Action Understanding (CVPR 2026 · similarity 0.89)
- Multi-Moments in Time: Learning and Interpreting Models for Multi-Action Video Understanding (TPAMI 2021 · similarity 0.89)
- Prototypical Calibrating Ambiguous Samples for Micro-Action Recognition (AAAI 2025 · similarity 0.89)
- UCF-STAR: A Large Scale Still Image Dataset for Understanding Human Actions (AAAI 2020 · similarity 0.89)
- Weakly-Supervised Temporal Action Detection for Fine-Grained Videos with Hierarchical Atomic Actions (ECCV 2022 · similarity 0.88)
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.
