Research map / Video / Action recognition
Semi-Supervised Temporal Action Detection with Proposal-Free Masking
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 |
|---|---|---|---|
| add Lc classification loss with class balancing (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 29.8 -> 32.1 | Improved |
| add feature reconstruction loss Lrec (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 30.2 -> 32.1 | Improved |
| add refinement loss Lref (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 30.7 -> 32.1 | Improved |
| add dice mask loss (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 31.2 -> 32.1 | Improved |
| add weighted CE mask loss (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 31.5 -> 32.1 | Improved |
| add pretext pre-training (mask, reconstruction, temporal position losses) (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 30.5 -> 32.1 | Improved |
| add pre-training and Lc (SSL modules) to use unlabeled data (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 28.3 -> 32.1 | Improved |
| add pre-training and Lc (SSL modules) (ablation) | SPOT · ActivityNet v1.3 (60% labels) | avg mAP: 34.0 -> 35.2 | Improved |
| add boundary refinement with both foreground and background terms (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 30.7 -> 32.1 | Improved |
| add positional encoding (learnable) (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 32.1 -> 29.4 | Hurt |
| add non-learnable positional encoding (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 32.1 -> 24.2 | Hurt |
| keep temporal ordering loss Ltp during fine-tuning (ablation) | SPOT · ActivityNet v1.3 (10% labels) | avg mAP: 32.1 -> 31.9 | Hurt |
Closest papers by meaning
- Proposal-Free Temporal Action Detection via Global Segmentation Mask Learning (ECCV 2022 · similarity 0.91)
- Dual DETRs for Multi-Label Temporal Action Detection (CVPR 2024 · similarity 0.88)
- End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames (CVPR 2024 · similarity 0.88)
- Stable Mean Teacher for Semi-supervised Video Action Detection (AAAI 2025 · similarity 0.88)
- HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding (NeurIPS 2026 · similarity 0.88)
- Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions (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.
