Research map / Video / Video-language understanding
Video Joint Modelling Based on Hierarchical Transformer for Co-summarization
Haopeng Li, Qiuhong Ke, Mingming Gong, Rui Zhang
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 hierarchical Transformer to the plain baseline (ablation) | baseline Transformer · SumMe canonical | F-measure: 0.485 -> 0.494 | Improved |
| add frame-level video joint modelling alone to the baseline (ablation) | baseline Transformer · SumMe canonical | F-measure: 0.485 -> 0.483 | Hurt |
| add Transformer-based reconstruction to the baseline (ablation) | baseline Transformer · SumMe canonical | F-measure: 0.485 -> 0.488 | Improved |
| add shot-level joint modelling on top of hierarchical Transformer (ablation) | VJMHT · SumMe canonical | F-measure: 0.496 -> 0.506 | Improved |
| add reconstruction on top of CS and HT (ablation) | VJMHT · TVSum transfer | F-measure: 0.571 -> 0.583 | Improved |
| add all three components to baseline (ablation) | VJMHT · TVSum transfer | F-measure: 0.561 -> 0.583 | Improved |
| use 2 videos rather than 1 for joint modelling (ablation) | VJMHT · TVSum | Kendall's tau: 0.088 -> 0.097 | Improved |
| use 4 videos rather than 2 for joint modelling (ablation) | VJMHT · TVSum | Kendall's tau: 0.097 -> 0.085 | Hurt |
| pick joint-modelling videos from the same cluster instead of random (ablation) | VJMHT · TVSum | Kendall's tau: 0.085 -> 0.097 | Improved |
| join random or inter-cluster videos versus none (ablation) | VJMHT · TVSum | Kendall's tau: 0.088 -> 0.085 | No clear effect |
| F-Transformer instead of C3D for frame-level modelling (ablation) | VJMHT · TVSum | F-measure: 0.572 -> 0.609 | Improved |
| increase shot embedding dimension from 256 to 512 (ablation) | VJMHT · SumMe | F-measure: 0.459 -> 0.494 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SumGraph — SumGraph: Video Summarization via Recursive Graph Modeling (ECCV 2020) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Video Summarization with Large Language Models (CVPR 2025 · similarity 0.87)
- Weakly-supervised Video Summarization using Variational Encoder-Decoder and Web Prior (ECCV 2018 · similarity 0.87)
- COOT: Cooperative Hierarchical Transformer for Video-Text Representation Learning (NeurIPS 2020 · similarity 0.87)
- CLIP-It! Language-Guided Video Summarization (NeurIPS 2021 · similarity 0.87)
- Query-Focused Video Summarization: Dataset, Evaluation, and a Memory Network Based Approach (CVPR 2017 · similarity 0.87)
- V2Xum-LLM: Cross-Modal Video Summarization with Temporal Prompt Instruction Tuning (AAAI 2025 · 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.
