Dataset.md

Holistic Coverage and Faithfulness Evaluation of Large Vision-Language Models

• For objects and relations, you need to download GQA dataset first. Next, you can extract the images with ids in:

  • ../human_annotation/object_existence.jsonl
  • ../human_annotation/relation_comparative.jsonl
  • ../human_annotation/relation_positional.jsonl

• For attributes, you can download the images here: ../datasets/attribute/download.txt