DOGR: A Multimodal Benchmark Dataset for Document and Chart Understanding

DOGR is a training and evaluation dataset for document-grounded reasoning, created by yinanzhou1 and last updated on September 19, 2025. The dataset appears to contain multimodal data, including PDFs, charts, and posters, as indicated by the separate archive files for ChartQA and Crello data. The description notes that obtaining the missing PDF images is a prerequisite for full use.

Use Cases

• Training models for document-grounded reasoning based on the described multimodal content.
• Benchmarking visual question answering systems on chart data from the ChartQA_PCG.zip archive.
• Evaluating model performance on poster and design elements from the Crello.zip archive.
• Developing PDF parsing and information extraction tools based on the dataset's PDF data component.

Strengths

• Includes a dedicated training and evaluation set, suggesting a structured benchmark purpose.
• Provides separate, focused data components for charts and posters, as indicated by the archive files.
• Dataset documentation details the directory structure and file purposes, as mentioned in the description.

Limitations

• Description metadata is limited; actual data quality, column definitions, and sample size require manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.

Provenance

Source

yinanzhou1 on Hugging Face

Collection Method

Method of data gathering is not described.

Time Range

Temporal coverage is not specified.

Freshness

Last updated 2025-09-19 07:41:31; freshness should be verified.

Geography

Spatial coverage is not specified.