le723z/

Vision-Reasoning-QA

Source Dataset: Derived from R1-Vision-Reasoning-Instructions with rigorous quality filtering. Key Processing Steps:

Data Filtering:

Removed all examples containing bounding box annotations Excluded pure image captioning tasks (difficult to verify)

Quality Control:

Answers exceeding 5 words underwent model-based verification Implemented special handling for LaTeX-formatted responses Verified answer consistency using sequence… See the full description on the dataset page: https://huggingface.co/datasets/le723z/Vision-Reasoning-QA.