Research map / Video / Action recognition
Rethinking the Faster R-CNN Architecture for Temporal Action Localization
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 |
|---|---|---|---|
| Align each anchor scale's receptive field using multi-tower dilated convolutions (ablation) | TAL-Net proposal generation · THUMOS'14 RGB proposals, AN=10 | AR: 9.4 -> 14.0 | Helped |
| Add temporal context around proposals (ablation) | TAL-Net proposal generation · THUMOS'14 RGB proposals, AN=10 | AR: 14.0 -> 15.1 | Helped |
| Add temporal context around proposals (ablation) | TAL-Net proposal generation · THUMOS'14 flow proposals, AN=10 | AR: 16.3 -> 17.4 | Helped |
| Add temporal context to action classification (ablation) | TAL-Net · THUMOS'14 RGB action localization, tIoU=0.5 | mAP: 28.5 -> 31.9 | Helped |
| Add temporal context to action classification (ablation) | TAL-Net · THUMOS'14 flow action localization, tIoU=0.5 | mAP: 37.4 -> 38.2 | Helped |
| Fuse RGB and flow logits late rather than concatenating features early (ablation) | TAL-Net · THUMOS'14 action localization, tIoU=0.5 | mAP: 40.8 -> 42.8 | Helped |
| Fuse RGB and flow logits late rather than concatenating features early (ablation) | TAL-Net · THUMOS'14 action localization, tIoU=0.1 | mAP: 60.5 -> 59.8 | Hurt |
| Use TAL-Net rather than Gao et al. | THUMOS'14 action localization · THUMOS'14, tIoU=0.5 | mAP: 31.0 -> 42.8 | Helped |
| Use TAL-Net rather than Dai et al. | ActivityNet v1.3 action localization · ActivityNet v1.3 val, tIoU=0.5 | mAP: 36.44 -> 38.23 | Helped |
| Use TAL-Net rather than Shou et al. | ActivityNet v1.3 action localization · ActivityNet v1.3 val, tIoU=0.5 | mAP: 45.30 -> 38.23 | Hurt |
Closest papers by meaning
- Temporal Action Localization in Untrimmed Videos via Multi-Stage CNNs (CVPR 2016 · similarity 0.91)
- Temporal Action Localization by Structured Maximal Sums (CVPR 2017 · similarity 0.91)
- CDC: Convolutional-De-Convolutional Networks for Precise Temporal Action Localization in Untrimmed Videos (CVPR 2017 · similarity 0.91)
- Temporal Action Localization with Cross Layer Task Decoupling and Refinement (AAAI 2025 · similarity 0.89)
- Realigning Confidence with Temporal Saliency Information for Point-Level Weakly-Supervised Temporal Action Localization (CVPR 2024 · similarity 0.88)
- Test-Time Zero-Shot Temporal Action Localization (CVPR 2024 · 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.