MaRs-VQA: A Dataset for Benchmarking Multimodal LLMs on Abstract Visual Reasoning
Description:
MaRs-VQA is annotated based on MaRs-IB. It contains 18 questionnaires, each of questionnaire contains 80 AVR questions. The human study of MaRs-IB is rigorous.
All questionnaires are under Attribution-NonCommercial 3.0 MaRs-IB LICENSE, which means it allows people to use the work, or adaptations of the work, for noncommercial purposes only, and only as long as they give… See the full description on the dataset page: https://huggingface.co/datasets/vcog/marsvqa.
