Research map / Video / Action recognition
Action Recognition From Single Timestamp Supervision in Untrimmed Videos
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 |
|---|---|---|---|
| iteratively update timestamp-initialized sampling distributions using classifier responses (ablation) | Ours with TS supervision · THUMOS 14 | top-1 accuracy: 32.69 -> 55.15 (h=0.50); 33.59 -> 56.42 (h=0.75) | Helped |
| iteratively update timestamp-initialized sampling distributions using classifier responses (ablation) | Ours with TS supervision · BEOID | top-1 accuracy: 71.62 -> 83.11 (h=0.50); 74.32 -> 83.11 (h=0.75) | Helped |
| iteratively update timestamp-initialized sampling distributions using classifier responses (ablation) | Ours with TS supervision · EPIC Kitchens | top-1 accuracy: 21.39 -> 25.35 (h=0.50); 20.73 -> 23.86 (h=0.75) | Helped |
| use curriculum selection of the top h fraction of sampled training frames instead of all frames (ablation) | Ours with TS supervision · BEOID | top-1 accuracy: 64.86 -> 71.62 before update (h=1 to h=0.50); 70.27 -> 83.11 after update | Helped |
| use curriculum selection of the top h fraction of sampled training frames instead of all frames (ablation) | Ours with TS supervision · EPIC Kitchens | top-1 accuracy: 23.55 -> 21.39 before update (h=1 to h=0.50); 24.17 -> 25.35 after update | Helped |
| use curriculum selection of fewer sampled frames instead of all sampled frames (ablation) | Ours with TS supervision · THUMOS 14 | top-1 accuracy: 63.41 -> 33.59 before update (h=1 to h=0.75); 63.53 -> 56.42 after update | Hurt |
| replace video-level supervision with updated single timestamp supervision | Ours (TS) versus Untrimmed Net · BEOID | top-1 accuracy: 28.37 -> 85.14 | Helped |
| replace video-level supervision with updated single timestamp supervision | Ours (TS) versus Untrimmed Net · EPIC Kitchens | top-1 accuracy: 2.20 -> 26.22 | Helped |
| use single timestamps rather than full temporal bounds | Ours · BEOID | top-1 accuracy: 85.14 (TS) and 88.51 (TS in GT) versus 87.83 (Full) | Mixed |
| use timestamps drawn from action boundaries rather than rough timestamps that may fall outside actions | Ours with timestamp supervision · EPIC Kitchens | top-1 accuracy: 26.22 -> 32.53 | Helped |
Closest papers by meaning
- Temporal Action Segmentation From Timestamp Supervision (CVPR 2021 · similarity 0.91)
- Action Sets: Weakly Supervised Action Segmentation Without Ordering Constraints (CVPR 2018 · similarity 0.90)
- Learning Transferable Self-Attentive Representations for Action Recognition in Untrimmed Videos with Weak Supervision (AAAI 2019 · similarity 0.90)
- Fast Temporal Activity Proposals for Efficient Detection of Human Actions in Untrimmed Videos (CVPR 2016 · similarity 0.90)
- Action-Agnostic Point-Level Supervision for Temporal Action Detection (AAAI 2025 · 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.