Re-export of the PubMedVision_Alignment_VQA subset from
FreedomIntelligence/PubMedVision
processed for easier downstream consumption.
Single-image rows only: rows with multiple images dropped (~22% of original)
9 rows with missing image files (upstream packaging gap; e.g. pmc_9_0.jpg is referenced but absent from images_*.zip) are also dropped
conversations expanded into separate question and… See the full description on the dataset page:
https://huggingface.co/datasets/mtybilly/PubMedVision-Alignment-VQA.