DVBench is a benchmark for evaluating multimodal large language models on data videos, a storytelling medium that combines dynamic charts with structured narratives.
Dataset Card
DVBench is a benchmark for evaluating multimodal large language models on data videos, a storytelling medium that combines dynamic charts with structured narratives. - 300 data videos - 1,000 human-verified question-answer pairs - questionid: unique question identifier - questiontype: EM, MCQsingle, MCQmultiple, or Openended - video: video identifier matching videos/.mp4 - dimension: evaluation dimension - question: question or subtitle-cloze context - answer: reference answer - distractor1, distractor2, distractor3: distractor options for single-choice and multiple-choice questions - charttype: chart category for visual questions in Animation, Chart Perception, and Chart Reasoning…
Excerpt from the card by Bomiao Wang, licensed cc-by-nc-sa-4.0.
Structure
default 1,000 rows
| Split | Rows | Size |
|---|---|---|
| test | 1,000 | 1.9 MB |
Details
- Repository
- BomiaoWang/DVBench
- Publisher
- Bomiao Wang
- Task category
- Visual question answering
- Tags
- video, chart, multimodal
- Size category
- Not stated by the source
- Languages
- en
- Revision
- 1fc763f40348a069a503cf0a594ae26bd609d83e
- Last updated
- 2026-09-22
Files
6 files, 7.8 GB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| requirements.txt | Data | 79 B | — |
| test.parquet | Data | 751.2 KB | 57a187d7e252 |
| videos.zip | Data | 7.8 GB | 3cc0b8a35388 |
| README.md | Documentation | 3.7 KB | — |
| dataset-statistics.png | Other | 551.4 KB | f55f27e5e115 |
| .gitattributes | Repository | 2.5 KB | — |
License and Download
- License
- cc-by-nc-sa-4.0
- Access
- No access gate
Released by Bomiao Wang through its official repository on Hugging Face.