Research map / Video / Action recognition
An Empirical Study of End-to-End 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 |
|---|---|---|---|
| Jointly train the video encoder and detection head instead of training only the head | TSM ResNet-50 + TadTR · THUMOS14 | average mAP: 34.04 -> 45.25 | Helped |
| Jointly train the video encoder and detection head instead of training only the head | TSM ResNet-50 + TadTR · ActivityNet v1.3 | average mAP: 32.76 -> 34.14 | Helped |
| Add image augmentation to end-to-end training (ablation) | TSM ResNet-50 + TadTR · THUMOS14 | average mAP: 39.49 -> 45.25 | Helped |
| Add random cropping (ablation) | TSM ResNet-50 + TadTR · THUMOS14 | average mAP: 39.49 -> 42.81 | Helped |
| Add horizontal flipping after cropping (ablation) | TSM ResNet-50 + TadTR · THUMOS14 | average mAP: 42.81 -> 43.90 | Helped |
| Add rotation and photometric distortion after cropping and flipping (ablation) | TSM ResNet-50 + TadTR · THUMOS14 | average mAP: 43.90 -> 45.25 | Helped |
| Add rotation and photometric distortion after cropping and flipping (ablation) | TSM ResNet-18 + TadTR · ActivityNet v1.3 | average mAP: 33.40 -> 33.42 | No clear effect |
| Replace TSM ResNet-50 with SlowFast ResNet-50 (ablation) | TadTR · THUMOS14 | average mAP: 45.3 -> 54.2 | Helped |
| Replace TSM ResNet-50 with SlowFast ResNet-50 (ablation) | TadTR · ActivityNet v1.3 | average mAP: 34.14 -> 35.10 | Helped |
| Increase batch size from 4 to 16 (ablation) | TSM ResNet-18 + TadTR · ActivityNet v1.3 | average mAP: 33.40 -> 33.25 | No clear effect |
| Remove multi-scale feature fusion between I3D stages four and five (ablation) | I3D + TadTR · THUMOS14 | mAP: 45.1 -> 42.1 | Hurt |
| Use fully convolutional rather than snippet-wise frame sampling (ablation) | TSM ResNet-18 + TadTR · THUMOS14 | mAP: 34.25 -> 36.12 | Helped |
Closest papers by meaning
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.93)
- TE-TAD: Towards Full End-to-End Temporal Action Detection via Time-Aligned Coordinate Expression (CVPR 2024 · similarity 0.90)
- Dual DETRs for Multi-Label Temporal Action Detection (CVPR 2024 · similarity 0.90)
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (CVPR 2024 · similarity 0.89)
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos (CVPR 2024 · similarity 0.89)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 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.