Research map / Video / Action recognition
G-TAD: Sub-Graph Localization for Temporal Action Detection
Check 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 semantic graph convolution to GCNeXt (ablation) | G-TAD · ActivityNet-1.3 validation | average mAP: 32.63 -> 34.09 | Helped |
| Add temporal graph convolution to GCNeXt (ablation) | G-TAD · ActivityNet-1.3 validation | average mAP: 33.67 -> 34.09 | Helped |
| Increase GCNeXt cardinality from 1 to 32 with both graph streams (ablation) | G-TAD · ActivityNet-1.3 validation | average mAP: 33.88 -> 34.09 | Helped |
| Add semantic feature concatenation in SGAlign (ablation) | G-TAD · ActivityNet-1.3 validation | average mAP: 33.89 -> 34.09 | Helped |
| Use sample-rescale alignment (ablation) | G-TAD SGAlign without semantic feature concatenation · ActivityNet-1.3 validation | average mAP: 33.78 -> 33.89 | Helped |
| Use the union of semantic edges from all GCNeXt blocks (ablation) | G-TAD SGAlign · ActivityNet-1.3 validation | average mAP: 34.09 -> 33.95 | Hurt |
| Replace FC localization layers with 2D convolution (ablation) | G-TAD · ActivityNet-1.3 validation | average mAP: 34.08; 34.10; 33.77 | No clear effect |
| Increase input window length from 128 to 256 (ablation) | G-TAD · THUMOS-14 validation | mAP at IoU@0.5: 38.70 -> 40.16 | Helped |
| Increase input window length from 256 to 512 (ablation) | G-TAD · THUMOS-14 validation | mAP at IoU@0.5: 40.16 -> 34.38 | Hurt |
Closest papers by meaning
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · similarity 0.91)
- ACGNet: Action Complement Graph Network for Weakly-Supervised Temporal Action Localization (AAAI 2022 · similarity 0.90)
- Gaussian Temporal Awareness Networks for Action Localization (CVPR 2019 · similarity 0.90)
- Scene-Aware Spatiotemporal Generalization: Towards Robust Temporal Action Detection Across Domains (AAAI 2026 · similarity 0.89)
- Graph Attention Based Proposal 3D ConvNets for Action Detection (AAAI 2020 · similarity 0.89)
- Temporal Action Localization by Structured Maximal Sums (CVPR 2017 · similarity 0.89)
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.