Research map / Label-efficient and robust learning / Domain adaptation and generalization
Cross-Category Video Highlight Detection via Set-Based Learning
Minghao Xu, Hang Wang, Bingbing Ni, Riheng Zhu, Zhenbang Sun, Changhu Wang
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 Transformer encoder T to set-based learning module (ablation) | SL-module · YouTube Highlights (average) | mAP: 0.651 -> 0.693 | Improved |
| add Transformer encoder T to set-based learning module (ablation) | SL-module · TVSum (average) | top-5 mAP: 0.712 -> 0.733 | Improved |
| add Transformer encoder T to set-based learning module (ablation) | SL-module · ActivityNet (average) | mAP: 0.759 -> 0.779 | Improved |
| add knowledge distillation between coarse and fine learners (ablation) | DL-VHD · ActivityNet sport->household | mAP: 0.778 -> 0.793 | Improved |
| add knowledge distillation between coarse and fine learners (ablation) | DL-VHD · YouTube Highlights surfing->skiing | mAP: 0.658 -> 0.686 | Improved |
| add fine-grained learner to coarse learner with distillation (ablation) | DL-VHD · YouTube Highlights surfing->skiing | mAP: 0.631 -> 0.686 | Improved |
| add coarse-grained learner on top of source-only (fine learner) (ablation) | DL-VHD · ActivityNet sport->household | mAP: 0.707 -> 0.742 | Improved |
| use full DL-VHD instead of target-oracle on gymnastics and skiing | DL-VHD · YouTube Highlights surfing->skiing | mAP: 0.661 -> 0.686 | Improved |
| use full DL-VHD instead of strongest UDA baseline | DL-VHD · ActivityNet sport->household | mAP: 0.765 -> 0.793 | Improved |
| use set-based learning module instead of MINI-Net | SL-module · TVSum (average) | top-5 mAP: 0.698 -> 0.733 | Improved |
| use set size N of at least 16 | SL-module · YouTube Highlights | mAP: N >= 16 | Improved |
| set trade-off lambda around 1.0 | DL-VHD · cross-category tasks | mAP: around 1.0 | Improved |
Closest papers by meaning
- Contrastive Learning for Unsupervised Video Highlight Detection (CVPR 2022 · similarity 0.92)
- Less Is More: Learning Highlight Detection From Video Duration (CVPR 2019 · similarity 0.91)
- Exploring Domain Incremental Video Highlights Detection with the LiveFood Benchmark (AAAI 2024 · similarity 0.87)
- Bridging the Gap: A Unified Video Comprehension Framework for Moment Retrieval and Highlight Detection (CVPR 2024 · similarity 0.87)
- Watch Video, Catch Keyword: Context-aware Keyword Attention for Moment Retrieval and Highlight Detection (AAAI 2025 · similarity 0.86)
- TR-DETR: Task-Reciprocal Transformer for Joint Moment Retrieval and Highlight Detection (AAAI 2024 · similarity 0.85)
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.
