Sign in to view source links and access this dataset
Description
VisCoR-55K is a high-quality dataset for visual reasoning spanning five categories: General, Reasoning, Math, Graph/Chart, and OCR. It contains original visual question-answer pairs, matched contrastive VQA pairs, and high-quality rationales synthesized by the VC-STaR framework. The dataset was authored by 5551z and last updated on Hugging Face in April 2026.
Use Cases
Training visual question answering models based on the 55K VQA samples.
Evaluating model reasoning faithfulness using the provided contrastive VQA pairs.
Training or fine-tuning models to generate rationales based on the synthesized rationale component.
Benchmarking performance across different visual reasoning categories like Math and Graph/Chart.