Research map / Video / Action recognition
Asynchronous Interaction Aggregation for Action Detection
Jiajun Tang, Jin Xia, Xinzhi Mu, Bo Pang, Cewu Lu
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 the memory (temporal) interaction block to person and object blocks (ablation) | AIA serial IA · AVA v2.2 val | mAP: 28.04 -> 29.26 | Improved |
| add the object interaction block to person and memory blocks (ablation) | AIA serial IA · AVA v2.2 val | mAP: 28.86 -> 29.26 | Improved |
| add the person interaction block to object and memory blocks (ablation) | AIA serial IA · AVA v2.2 val | mAP: 28.92 -> 29.26 | Improved |
| add all three interaction types to the backbone (ablation) | SlowFast R-50 · AVA v2.2 val | mAP: 26.54 -> 29.26 | Improved |
| use 2 blocks per interaction type instead of 1 (ablation) | AIA serial IA · AVA v2.2 val | mAP: 29.26 -> 29.64 | Improved |
| use 3 blocks per interaction type instead of 2 (ablation) | AIA serial IA · AVA v2.2 val | mAP: 29.64 -> 29.61 | No clear effect |
| change the order of interaction blocks (e.g. M->P->O instead of P->O->M) (ablation) | AIA serial IA · AVA v2.2 val | mAP: 29.44 -> 29.64 | No clear effect |
| use serial instead of parallel aggregation of interaction blocks (ablation) | AIA IA structure · AVA v2.2 val | mAP: 28.85 -> 29.64 | Improved |
| add dense learnable connections between blocks (dense serial) (ablation) | AIA serial IA · AVA v2.2 val | mAP: 29.64 -> 29.80 | Improved |
| add asynchronous memory update instead of frozen memory features (ablation) | AIA IA · AVA v2.2 val | mAP: 28.07 -> 29.64 | Improved |
| add asynchronous memory update to LFB (ablation) | LFB (SlowFast re-implementation) · AVA v2.2 val | mAP: 27.02 -> 28.57 | Improved |
| use IA (P and O blocks) instead of a non-local branch (ablation) | SlowFast R-50 baseline · AVA v2.2 val | mAP: 26.85 -> 28.23 | Improved |
Closest papers by meaning
- Learning Asynchronous and Sparse Human-Object Interaction in Videos (CVPR 2021 · similarity 0.86)
- DCAN: Improving Temporal Action Detection via Dual Context Aggregation (AAAI 2022 · similarity 0.86)
- AOE-Net: Entities Interactions Modeling with Adaptive Attention Mechanism for Temporal Action Proposals Generation (IJCV 2022 · similarity 0.86)
- TEA: Temporal Excitation and Aggregation for Action Recognition (CVPR 2020 · similarity 0.85)
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.84)
- Action-Agnostic Point-Level Supervision for Temporal Action Detection (AAAI 2025 · similarity 0.83)
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.
