Research map / Video / Action recognition
Post-Processing Temporal Action Detection
Sauradip Nag, Xiatian Zhu, Yi-Zhe Song, Tao Xiang
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 |
|---|---|---|---|
| Apply GAP post-processing to existing TAD models | TAD models · ActivityNet-v1.3 and THUMOS14 | average mAP: +0.2 to +0.7 percentage points on ActivityNet; +0.2 to +0.5 percentage points on THUMOS | Improved |
| Apply GAP post-processing | BMN · ActivityNet-v1.3, temporal resolution 25 | average mAP: 28.1 -> 28.5 | Improved |
| Reduce BMN temporal resolution from 400 to 25 snippets (ablation) | BMN · ActivityNet-v1.3 | average mAP: 34.0 -> 28.1 | Hurt |
| Add temporal distribution smoothing (ablation) | BMN + GAP · ActivityNet-v1.3 | average mAP: 34.0 -> 34.3 | Improved |
| Add ground-truth calibration (ablation) | BMN without GAP · ActivityNet-v1.3 | average mAP: 33.9 -> 34.0 | Improved |
| Add ground-truth calibration (ablation) | BMN + GAP · ActivityNet-v1.3 | average mAP: 34.2 -> 34.3 | Improved |
| Use ceiling rather than floor quantization (ablation) | BMN + GAP ground-truth calibration · ActivityNet-v1.3 | average mAP: 34.3 -> 34.0 | Hurt |
| Integrate GAP during training (ablation) | RTDNet · ActivityNet-v1.3 | average mAP: 30.8 -> 31.4 | Improved |
| Add GAP inference to GAP-assisted training (ablation) | RTDNet · ActivityNet-v1.3 | average mAP: 31.4 -> 31.5 | Improved |
| Integrate GAP during training (ablation) | AFSD · ActivityNet-v1.3 | average mAP: 34.4 -> 35.0 | Improved |
| Add GAP inference to GAP-assisted training (ablation) | AFSD · ActivityNet-v1.3 | average mAP: 35.0 -> 35.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ActionFormer — ActionFormer: Localizing Moments of Actions with Transformers (ECCV 2022) · 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
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.91)
- Temporal Action Detection Model Compression by Progressive Block Drop (CVPR 2025 · similarity 0.89)
- Dual DETRs for Multi-Label Temporal Action Detection (CVPR 2024 · similarity 0.89)
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (CVPR 2024 · similarity 0.89)
- TE-TAD: Towards Full End-to-End Temporal Action Detection via Time-Aligned Coordinate Expression (CVPR 2024 · similarity 0.89)
- HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding (NeurIPS 2026 · similarity 0.87)
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.
