Research map / Video / Action recognition
Graph Convolutional Module for Temporal Action Localization in Videos
Runhao Zeng, Wenbing Huang, Mingkui Tan, Yu Rong, Peilin Zhao, Junzhou Huang, Chuang Gan
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 GCM to SSN (BSN proposals, I3D) | SSN · THUMOS14 tIoU=0.5 | mAP: 49.3 -> 50.9 | Improved |
| add GCM to R-C3D | R-C3D · THUMOS14 tIoU=0.5 | mAP: 28.90 -> 30.85 | Improved |
| add GCM to CBR | CBR · THUMOS14 tIoU=0.5 | mAP: 31.00 -> 32.24 | Improved |
| add GCM to one-stage D-SSAD | D-SSAD · THUMOS14 tIoU=0.5 | mAP: 43.21 -> 44.77 | Improved |
| add graph relations (GCM) in place of MLP for both classification and regression branches (ablation) | SSN · THUMOS14 tIoU=0.5 RGB / Flow | mAP: 36.82 -> 39.38 (RGB); 46.74 -> 48.76 (Flow) | Improved |
| use graph convolution instead of mean pooling over adjacent units (ablation) | SSN · THUMOS14 tIoU=0.5 RGB / Flow | mAP: 37.12 -> 39.38 (RGB); 46.96 -> 48.76 (Flow) | Improved |
| add self-addition of node feature (ablation) | GCM · THUMOS14 tIoU=0.5 RGB / Flow | mAP: 38.32 -> 39.38 (RGB); 47.62 -> 48.76 (Flow) | Improved |
| embed features before cosine edge weight (embed-cos-sim) (ablation) | GCM · THUMOS14 tIoU=0.5 | mAP: 39.38 -> 39.27 (RGB); 48.76 -> 48.92 (Flow) | No clear effect |
| use learned attention weights on a fully-connected graph instead of GCM graph (ablation) | SSN · THUMOS14 tIoU=0.5 | mAP: 50.9 -> 49.7 | Hurt |
| add surrounding edges (ablation) | GCM · THUMOS14 tIoU=0.5 RGB / Flow | mAP: 38.80 -> 39.38 (RGB); 47.69 -> 48.76 (Flow) | Improved |
| add contextual edges (ablation) | GCM · THUMOS14 tIoU=0.5 RGB / Flow | mAP: 38.28 -> 39.38 (RGB); 47.57 -> 48.76 (Flow) | Improved |
| add semantic edges (ablation) | GCM · THUMOS14 tIoU=0.5 RGB / Flow | mAP: 39.02 -> 39.38 (RGB); 47.38 -> 48.76 (Flow) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- R-C3D — R-C3D: Region Convolutional 3D Network for Temporal Activity Detection (ICCV 2017) · better on 2 of 2 shared results
- GTAD* — G-TAD: Sub-Graph Localization for Temporal Action Detection (CVPR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Graph Convolutional Networks for Temporal Action Localization (ICCV 2019 · similarity 0.95)
- G-TAD: Sub-Graph Localization for Temporal Action Detection (CVPR 2020 · similarity 0.95)
- Temporal Action Localization in the Deep Learning Era: A Survey (TPAMI 2023 · similarity 0.95)
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · similarity 0.93)
- Video Self-Stitching Graph Network for Temporal Action Localization (ICCV 2021 · similarity 0.93)
- One-Shot Action Localization by Learning Sequence Matching Network (CVPR 2018 · similarity 0.92)
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.
