train.jsonl & val.jsonl

Images, questions & answers for the train and val split in the following format:

• question_id: [int] ID of the question
• question: [str] Content of the question
• image_id: [str] ID of the image
• image_name: [str] Name of the image in CV dataset
• image_path: [str] URL or path to the image
• source: [str] Name of the image source
• answers: [str array] List of answers
• correct: [int array] List of 1 or 0 indicating the correctness of each answer
• categories: [int array] List of categories

train_ann.jsonl & val_ann.jsonl

Annotations for the train and val split in the following format:

• question_id: [int] ID of the question
• hop: [int] Number of hops
• property_id: [str] Wikidata ID of the property used to construct the question
• property_label: [str] Wikidata name of the property used to construct the question
• entity_id: [str] Wikidata ID of the entity used to construct the question
• entity_label: [str] Wikidata name of the entity used to construct the question
• route: [str array] List of nodes from the image to the question
• has_scene_graph: [bool] Indicates if the question includes a scene graph information

categories.json

Contains the list of categories