Research map / Video / Video-language understanding
Global-and-Local Relative Position Embedding for Unsupervised Video Summarization
Yunjae Jung, Donghyeon Cho, Sanghyun Woo, In So Kweon
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 self-attention embedding (SAE) (ablation) | CSNet · TVSum | Kendall's tau / Spearman's rho: 0.025 and 0.034 to 0.034 and 0.045 | Improved |
| add relative position embedding (RPE) (ablation) | CSNet · TVSum | Kendall's tau / Spearman's rho: 0.025 and 0.033 to 0.034 and 0.044 | Improved |
| add RPE on top of SAE (without decomposition) (ablation) | CSNet SAE · TVSum | Kendall's tau: 0.034 -> 0.033 | No clear effect |
| add global-and-local decomposition with 8 segments to RPE (ablation) | CSNet+RPE · TVSum | Kendall's tau: 0.033 -> 0.070 | Improved |
| use 8 segments instead of 2 for GL-RPE (ablation) | CSNet+RPE+GL · TVSum | Kendall's tau: 0.039 -> 0.070 | Improved |
| increase segments from 8 to 10 for RPE+GL (ablation) | CSNet+RPE+GL · TVSum | Kendall's tau: 0.070 -> 0.066 | Hurt |
| use both global and local sampling instead of global only (RPE, 8 segments) (ablation) | CSNet+RPE · TVSum | Kendall's tau: 0.064 -> 0.070 | Improved |
| use both global and local instead of global only (SAE, 8 segments) (ablation) | CSNet+SAE · TVSum | Kendall's tau: 0.065 -> 0.066 | No clear effect |
| add GL+RPE to GAN backbone | GAN · TVSum | Kendall's tau: 0.024 -> 0.064 | Improved |
| add GL+RPE to CSNet backbone | CSNet · TVSum | Kendall's tau: 0.025 -> 0.070 | Improved |
| add GL+RPE to CSNet | CSNet · SumMe (Canonical) | F-score (%): 51.3 -> 50.2 | Hurt |
| add GL+RPE to CSNet | CSNet · TVSum (Canonical) | F-score (%): 58.8 -> 59.1 | Improved |
Closest papers by meaning
- Enhancing Video Summarization via Vision-Language Embedding (CVPR 2017 · similarity 0.88)
- Retrospective Encoders for Video Summarization (ECCV 2018 · similarity 0.88)
- Video Summarization with Long Short-Term Memory (ECCV 2016 · similarity 0.87)
- Query-Focused Video Summarization: Dataset, Evaluation, and a Memory Network Based Approach (CVPR 2017 · similarity 0.87)
- Video Summarization with Large Language Models (CVPR 2025 · similarity 0.85)
- Scaling Up Video Summarization Pretraining with Large Language Models (CVPR 2024 · 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.
