Research map / Video / Video segmentation and tracking
Less Is More: Learning Highlight Detection From Video Duration
Bo Xiong, Yannis Kalantidis, Deepti Ghadiyaram, Kristen Grauman
Check 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 |
|---|---|---|---|
| Use domain-specific rather than domain-agnostic Instagram training | Duration-based highlight detector · YouTube Highlights | mAP: 0.505 -> 0.564 | Improved |
| Use domain-specific rather than domain-agnostic Instagram training | Duration-based highlight detector · TVSum | Top-5 mAP: 0.524 -> 0.563 | Improved |
| Add learned latent weighting of noisy ranking pairs (ablation) | Duration-based highlight detector · YouTube Highlights | mAP: 0.425 -> 0.564 | Improved |
| Add learned latent weighting of noisy ranking pairs (ablation) | Duration-based highlight detector · TVSum | mAP: 0.400 -> 0.563 | Improved |
| Replace learned, continuous input-conditioned pair weights with alternating binary latent selection (ablation) | Duration-based highlight detector · YouTube Highlights | mAP: 0.458 -> 0.564 | Improved |
| Replace learned, continuous input-conditioned pair weights with alternating binary latent selection (ablation) | Duration-based highlight detector · TVSum | mAP: 0.444 -> 0.563 | Improved |
| Increase training videos per domain from 1,000 to 10,000 (ablation) | Ours-S · YouTube Highlights | mAP: +6.5% | Improved |
| Increase training videos per domain from 1,000 to 10,000 (ablation) | Ours-A · YouTube Highlights | mAP: +3.7% | Improved |
| Use group size n=8 for pair-weight softmax (ablation) | Duration-based highlight detector · YouTube Highlights and TVSum | mAP | Improved |
| Vary short- and long-video duration cutoffs within the tested range (ablation) | Duration-based highlight detector · YouTube Highlights and TVSum | mAP | No clear effect |
| Weight test pairs with the learned latent pair predictor (ablation) | Duration-based ranking model · Held-out short-versus-long video segment pairs | successful ranking rate: 58.2% -> 87.2% | Improved |
| Train on YouTube8M instead of Instagram videos | Ours-A · YouTube Highlights | mAP: 0.505 -> 0.317 | Hurt |
Closest papers by meaning
- Contrastive Learning for Unsupervised Video Highlight Detection (CVPR 2022 · similarity 0.92)
- Cross-Category Video Highlight Detection via Set-Based Learning (ICCV 2021 · similarity 0.91)
- Watch Video, Catch Keyword: Context-aware Keyword Attention for Moment Retrieval and Highlight Detection (AAAI 2025 · similarity 0.85)
- Exploring Domain Incremental Video Highlights Detection with the LiveFood Benchmark (AAAI 2024 · similarity 0.85)
- Bridging the Gap: A Unified Video Comprehension Framework for Moment Retrieval and Highlight Detection (CVPR 2024 · similarity 0.85)
- Video Repurposing from User Generated Content: A Large-scale Dataset and Benchmark (AAAI 2025 · similarity 0.84)
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.
