This is a corrected version of the MMVP benchmark, re-hosted by lmms-lab-eval for use with lmms-eval.
The original MMVP/MMVP dataset was uploaded in imagefolder format, which only exposes the image column. The text annotations (Question, Options, Correct Answer, Index) from the accompanying Questions.csv were not loaded into the dataset, making it unusable for evaluation.
This version reconstructs the complete… See the full description on the dataset page:
https://huggingface.co/datasets/lmms-lab-eval/MMVP.
