73,702 annotated samples link chart-based answers to supporting visual elements via bounding boxes. This benchmark extends ChartQA and TinyChart-PoT datasets, created by author 'omoured' and last updated on July 9, 2025. It includes 55,789 training, 6,223 validation, and 11,690 testing instances.
Use Cases
- Train visual grounding models based on explicit bounding box annotations linking answers to chart elements.
- Benchmark chart-based question answering systems based on the large-scale, annotated test set.
- Evaluate model performance on visual reasoning tasks based on the structured validation and test splits.
Strengths
- Large-scale benchmark with 73,702 total annotated samples.
- Explicit bounding box annotations provide a direct link between answers and visual chart elements.
- Structured splits with 55,789 training, 6,223 validation, and 11,690 testing instances.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- Author 'omoured' on Hugging Face.
- Collection Method
- Extends the ChartQA and TinyChart-PoT datasets by adding bounding box annotations.
- Time Range
- null
- Freshness
- Last updated 2025-07-09 07:54:47; freshness should be verified.
- Geography
- null