Research map / Video / Action recognition
Graph Attention Based Proposal 3D ConvNets for Action Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 intra+inter attention GCN (AGCN) | R-C3D · THUMOS 2014 | mAP@IoU0.5 (%): 34.8 -> 38.4 | Helped |
| add framewise classifier on top of AGCN (ablation) | AGCN-P-C3D · THUMOS 2014 | average mAP (%): 43.5 -> 44.6 | Helped |
| add framewise classifier and AGCN | R-C3D · THUMOS 2014 | average mAP (%): 40.9 -> 44.6 | Helped |
| apply to ResNet34 backbone | P-ResNet34 · THUMOS 2014 | mAP@IoU0.5 (%): 36.9 -> 38.2 | Helped |
| add attention to GCN (vs 1-layer plain GCN) (ablation) | P-C3D · THUMOS 2014 | mAP@IoU0.5 (%): 34.7 -> 38.4 | Helped |
| attention AGCN vs 3-layer plain GCN (ablation) | P-C3D · THUMOS 2014 | mAP@IoU0.5 (%): 34.8 -> 38.4 | Helped |
| AGCN vs GAT-style attention (ablation) | P-C3D · THUMOS 2014 | mAP@IoU0.5 (%): 34.4 -> 38.4 | Helped |
| average rows (favg_r) instead of learnable parameter matrix in intra attention (ablation) | AGCN-P-C3D · THUMOS 2014 | mAP@IoU0.5 (%): 33.9 -> 38.4 | Helped |
| add fmask in inter attention (ablation) | AGCN-P-C3D · THUMOS 2014 | mAP@IoU0.5 (%): 35.8 -> 38.4 | Helped |
| add fmask in inter attention (ablation) | AGCN-P-C3D · THUMOS 2014 | mAP@IoU0.1 (%): 58.7 -> 57.2 | Hurt |
| inter attention only (no intra) (ablation) | P-C3D · THUMOS 2014 | mAP | Hurt |
| naive fusion of Ainter and Xintra (ablation) | P-C3D · THUMOS 2014 | mAP | Mixed |
Closest papers by meaning
- G-TAD: Sub-Graph Localization for Temporal Action Detection (CVPR 2020 · similarity 0.89)
- A Structured Model for Action Detection (CVPR 2019 · similarity 0.89)
- Learnable Graph Convolutional Attention Networks (ICLR 2023 · similarity 0.87)
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.87)
- Are Spatial-Temporal Graph Convolution Networks for Human Action Recognition Over-Parameterized? (CVPR 2025 · similarity 0.86)
- GCNext: Towards the Unity of Graph Convolutions for Human Motion Prediction (AAAI 2024 · similarity 0.86)
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.