Research map / Vision-language and multimodal / Visual question answering
Towards Comprehensive Lecture Slides Understanding: Large-scale Dataset and Effective Method
Enming Zhang, Yuzhe Li, Yuliang Liu, Yingying Zhu, Xiang Bai
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 Text Relation Prediction Module (RPH, RAL, pseudo labels) (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | ROUGE-1: 36.3 -> 42.2 | Improved |
| add Text Relation Prediction Module (RPH, RAL, pseudo labels) (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | METEOR: 26.9 -> 31.3 | Improved |
| add weak supervision pseudo labels to RPH+RAL (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | ROUGE-1: 37.5 -> 42.2 | Improved |
| learn relations with RPH under weak supervision instead of feeding pseudo labels directly to RAL (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | ROUGE-1: 41.8 -> 42.2 | Improved |
| use DL-Cluster pseudo labels instead of Kmeans k=3 (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | ROUGE-1: 40.9 -> 42.2 | Improved |
| use DL-Cluster pseudo labels instead of DBSCAN (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | METEOR: 30.6 -> 31.3 | Improved |
| use DL-Cluster instead of HAC (n=3) (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | METEOR: 30.1 -> 31.3 | Improved |
| add spatial relation to semantic-only DL-Cluster (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | ROUGE-1: 38.3 -> 42.2 | Improved |
| add semantic relation to spatial-only DL-Cluster (ablation) | SlideParser · LecSlides-370K Lecture Summary (val) | ROUGE-1: 41.4 -> 42.2 | Improved |
| pretrain on LecSlides-370K before fine-tuning on SlideVQA | SlideParser · SlideVQA test | EM: 37.5 -> 42.4 | Improved |
| pretrain on LecSlides-370K before fine-tuning on SlideVQA | SlideParser · SlideVQA test | F1: 43.9 -> 47.7 | Improved |
| use SlideParser over UDOP on lecture summary | SlideParser vs UDOP · LecSlides-370K Lecture Summary (test) | ROUGE-1: 35.3 -> 41.7 | Improved |
Closest papers by meaning
- Multi-type MOOCs Recommendation: Leveraging Deep Multi-Relational Representation and Hierarchical Reasoning (AAAI 2025 · similarity 0.81)
- QuerySum: A Multi-Document Query-Focused Summarization Dataset Augmented with Similar Query Clusters (AAAI 2024 · similarity 0.81)
- GRIP: A Graph-Based Reasoning Instruction Producer (NeurIPS 2025 · similarity 0.80)
- SlideChat: A Large Vision-Language Assistant for Whole-Slide Pathology Image Understanding (CVPR 2025 · similarity 0.80)
- K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs (NeurIPS 2026 · similarity 0.80)
- OpenWebMath: An Open Dataset of High-Quality Mathematical Web Text (ICLR 2024 · similarity 0.80)
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.
