Research map / Video / Action recognition
CycleACR: Cycle Modeling of Actor-Context Relations for Video Action Detection
Lei Chen, Tong Zhan, Yibing Song, Gangshan Wu, Limin 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 |
|---|---|---|---|
| add actor-to-context reorganization to context-to-actor enhancement (cycle) (ablation) | CycleACR · AVA v2.2 (SlowFast R101) | mAP: 33.47 -> 34.12 | Improved |
| use actor-to-context instead of normal context-to-actor attention (ablation) | cross-attention · AVA v2.2 (SlowFast R101) | mAP: 33.47 -> 33.59 | Improved |
| add local and global branches (ablation) | CycleACR (no memory bank) · AVA v2.2 (SlowFast R101) | mAP: 29.78 -> 32.57 | Improved |
| add local branch only (ablation) | CycleACR (no memory bank) · AVA v2.2 (SlowFast R101) | mAP: 29.78 -> 31.82 | Improved |
| add memory bank to local and global branches (ablation) | CycleACR · AVA v2.2 (SlowFast R101) | mAP: 32.57 -> 34.12 | Improved |
| add global branch to local branch with memory bank (ablation) | CycleACR · AVA v2.2 (SlowFast R101) | mAP: 33.40 -> 34.12 | Improved |
| use non-local instead of transformer in A2C-R (ablation) | A2C-R · AVA v2.2 (SlowFast R101) | mAP: 33.25 -> 33.40 | Improved |
| use attention instead of averaging for C2A-E (ablation) | C2A-E · AVA v2.2 (SlowFast R101) | mAP: 32.88 -> 33.40 | Improved |
| stack 3 layers instead of 2 (ablation) | CycleACR · AVA v2.2 (SlowFast R101) | mAP: 34.12 -> 33.54 | Hurt |
| use 2 stacked layers instead of 1 (ablation) | CycleACR · AVA v2.2 (SlowFast R101) | mAP: 33.96 -> 34.12 | Improved |
| add CycleACR to SlowFast R50 | backbone · AVA v2.2 | mAP: +3.6 | Improved |
| add CycleACR to SlowFast R101 | backbone · AVA v2.2 | mAP: +4.3 | Improved |
Closest papers by meaning
- Actor-Context-Actor Relation Network for Spatio-Temporal Action Localization (CVPR 2021 · similarity 0.89)
- Action Genome: Actions As Compositions of Spatio-Temporal Scene Graphs (CVPR 2020 · similarity 0.88)
- Context-Aware RCNN: A Baseline for Action Detection in Videos (ECCV 2020 · similarity 0.88)
- Unified Graph Structured Models for Video Understanding (ICCV 2021 · similarity 0.88)
- Open-Vocabulary Video Relation Extraction (AAAI 2024 · similarity 0.87)
- Learning Actor Relation Graphs for Group Activity Recognition (CVPR 2019 · similarity 0.87)
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.
