Research map / Video / Action recognition
Actor-Context-Actor Relation Network for Spatio-Temporal Action Localization
Junting Pan, Siyu Chen, Mike Zheng Shou, Yu Liu, Jing Shao, Hongsheng Li
Check 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 high-order actor-context-actor reasoning (HR²O) (ablation) | SlowFast R-50 baseline · AVA v2.2 | mAP: 25.39 -> 27.83 | Improved |
| Add Actor-Context Feature Bank (ablation) | HR²O · AVA v2.2 | mAP: 27.83 -> 28.84 | Improved |
| Replace ACFB with Long-term Feature Bank (ablation) | HR²O · AVA v2.2 | mAP: 27.83 -> 27.75 | Hurt |
| Use modified non-local attention rather than average aggregation (ablation) | HR²O · AVA v2.2 | mAP: 26.97 -> 27.83 | Improved |
| Use modified non-local attention rather than a relation network (ablation) | HR²O · AVA v2.2 | mAP: 27.18 -> 27.83 | Improved |
| Encode actor-context relations before actor-actor reasoning (ablation) | ACAR relation reasoning · AVA v2.2 | mAP: 27.62 -> 27.83 | Improved |
| Increase HR²O depth from one to two non-local blocks (ablation) | HR²O · AVA v2.2 | mAP: 27.63 -> 27.83 | Improved |
| Increase HR²O depth from two to three non-local blocks (ablation) | HR²O · AVA v2.2 | mAP: 27.83 -> 27.25 | Hurt |
| Add actor-context-actor relation modeling instead of only first-order relation methods (ablation) | SlowFast R-50 baseline · AVA v2.2 | mAP: 26.10 -> 27.83 | Improved |
| Use ACAR without ACFB rather than the SlowFast baseline | SlowFast R-50 · UCF101-24 split 1 | frame-mAP: 82.4 -> 84.3 | Improved |
| Add ACAR modeling | R-101 SlowFast model · AVA-Kinetics val | mAP: +2.86 mAP | Improved |
| Use ACAR-Net rather than the SlowFast R-101 baseline | R-101 SlowFast model · AVA-Kinetics val | mAP: 32.98 -> 35.84 | Improved |
Closest papers by meaning
- Actor-centric Relation Network (ECCV 2018 · similarity 0.92)
- Graph Convolutional Module for Temporal Action Localization in Videos (TPAMI 2021 · similarity 0.89)
- Graph Convolutional Networks for Temporal Action Localization (ICCV 2019 · similarity 0.89)
- Enriching Local and Global Contexts for Temporal Action Localization (ICCV 2021 · similarity 0.89)
- CycleACR: Cycle Modeling of Actor-Context Relations for Video Action Detection (TPAMI 2025 · similarity 0.89)
- Bilateral Relation Distillation for Weakly Supervised Temporal Action Localization (TPAMI 2023 · 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.
