README.md

metadata

viewer: false

Overview

• Vision question Answer (VQA) dataset: VQA is a new dataset containing open-ended questions about images. These questions require an understanding of vision, language and commonsense knowledge to answer.
• Reference: Split into small-train & small-val dataset from https://huggingface.co/datasets/Graphcore/vqa validation dataset

Dataset Structure

Data Instances

A data point comprises an image and its object annotations.

{'question': 'Where is he looking?',
 'question_type': 'none of the above',
 'question_id': 262148000,
 'image_id': 'images/COCO_val2014_000000262148.jpg',
 'answer_type': 'other',
 'label': {'ids': ['at table', 'down', 'skateboard', 'table'],
  'weights': [0.30000001192092896,
   1.0,
   0.30000001192092896,
   0.30000001192092896]}}

Data Fields

• question: the question to be answered from the image
• question_type:
• image_id: the path to the image the question refers to
• answer_type:
• label: the annotations
  • ids:
  • weights:

Data Splits

• Training dataset (1169)
• Val dataset (100)

Usage

from datasets import load_dataset
dataset = load_dataset("SIS-2024-spring/coco_vqa_small_dataset")