Research map / Video / Action recognition
Video Action Recognition with Attentive Semantic Units
Yifei Chen, Dapeng Chen, Ruijin Liu, Hao Li, Wei Peng
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 transformer with semantic-unit-queried cross-attention decoder (ablation) | ASU-B/16 · Kinetics-400 | top-1 accuracy: 81.1 -> 82.0 | Improved |
| add region-level representation (MRA) (ablation) | ASU-B/16 · Kinetics-400 | top-1 accuracy: 81.1 -> 81.8 | Improved |
| combine region-level representation with semantic units (ablation) | ASU-B/16 · Kinetics-400 | top-1 accuracy: 82.0 -> 82.4 | Improved |
| add 1D temporal convolution in video decoder (ablation) | ASU-B/16 · Kinetics-400 | top-1 accuracy: 82.4 -> 82.6 | Improved |
| add semantic units as decoder queries (ablation) | ASU-B/16 · HMDB-51 2-shot | top-1 accuracy: 52.6 -> 57.7 | Improved |
| add semantic units as decoder queries (ablation) | ASU-B/16 · UCF-101 zero-shot | top-1 accuracy: 64.9 -> 75.0 | Improved |
| use only one type of semantic unit instead of all four (ablation) | ASU-B/16 · UCF-101 zero-shot | top-1 accuracy: 75.0 -> 73.8 (objects only) | Hurt |
| increase regions from 4 to 8 using 4x2 split (ablation) | ASU-B/16 · Kinetics-400 | top-1 accuracy: 82.6 -> 81.8 | Hurt |
| use vertical divisions of regions (2x2) (ablation) | ASU-B/16 · Kinetics-400 | top-1 accuracy: 82.6 -> 82.2 | Hurt |
| use learnable prompts (CoOp, CoCoOp) instead of fixed template (ablation) | ASU-B/16 · Kinetics-400 | top-1 accuracy: 82.6 -> 82.3 (CoOp), 82.2 (CoCoOp) | No clear effect |
| apply ASU in zero-shot on HMDB-51 and UCF-101 vs X-CLIP-B/16 | X-CLIP-B/16 · HMDB-51 / UCF-101 zero-shot | top-1 accuracy: 44.6 -> 48.1; 72.0 -> 75.0 | Improved |
Closest papers by meaning
- Align Before Adapt: Leveraging Entity-to-Region Alignments for Generalizable Video Action Recognition (CVPR 2024 · similarity 0.89)
- Action Detail Matters: Refining Video Recognition with Local Action Queries (CVPR 2025 · similarity 0.88)
- OST: Refining Text Knowledge with Optimal Spatio-Temporal Descriptor for General Video Recognition (CVPR 2024 · similarity 0.88)
- Weakly Supervised Temporal Action Localization via Dual-Prior Collaborative Learning Guided by Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- LLaVAction: evaluating and training multi-modal large language models for action understanding (ICLR 2026 · similarity 0.88)
- SeFAR: Semi-supervised Fine-grained Action Recognition with Temporal Perturbation and Learning Stabilization (AAAI 2025 · 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.
