VQA v2.0: Visual Question Answering Dataset

# VQA v2.0 - Visual Question Answering
# Based on Goyal et al., CVPR 2017
# Paper: https://arxiv.org/abs/1612.00837
# Dataset: https://visualqa.org/
#
# This task presents an image and a natural language question about it.
# Annotators provide a free-form answer and, for yes/no questions,
# also select from a radio button group.
#
# Question Types:
# - yes/no: Questions that can be answered with yes or no
# - number: Questions asking about counts or quantities
# - other: Open-ended questions requiring descriptive answers
#
# Annotation Guidelines:
# 1. Look at the image carefully
# 2. Read the question
# 3. Provide a concise, accurate answer in the text field
# 4. For yes/no questions, also select the appropriate radio button
# 5. Answer based only on what you can see in the image

annotation_task_name: "VQA v2.0 - Visual Question Answering"
task_dir: "."

data_files:
  - sample-data.json

item_properties:
  id_key: "id"
  text_key: "text"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

port: 8000
server_name: localhost

annotation_schemes:
  # Step 1: Free-form answer
  - annotation_type: text
    name: answer
    description: "Provide a concise answer to the question about the image"

  # Step 2: Yes/No/NA for applicable questions
  - annotation_type: radio
    name: yes_no_answer
    description: "For yes/no questions, select the appropriate answer. For other question types, select Not Applicable."
    labels:
      - "Yes"
      - "No"
      - "Not Applicable"
    keyboard_shortcuts:
      "Yes": "1"
      "No": "2"
      "Not Applicable": "3"
    tooltips:
      "Yes": "The answer to the yes/no question is yes"
      "No": "The answer to the yes/no question is no"
      "Not Applicable": "The question is not a yes/no question"

annotation_instructions: |
  You will be shown an image and a question about it. Your task is to:
  1. Study the image carefully.
  2. Read the question.
  3. Type a concise answer in the text field (1-3 words preferred).
  4. If the question is a yes/no question, also select Yes or No. Otherwise, select Not Applicable.

  Answer based only on what you can see in the image. Keep answers brief and specific.

html_layout: |
  <div style="padding: 15px; max-width: 800px; margin: auto;">
    <div style="background: #f8fafc; border: 1px solid #e2e8f0; border-radius: 8px; padding: 16px; margin-bottom: 16px; text-align: center;">
      <img src="{{image_url}}" style="max-width: 100%; max-height: 500px; border-radius: 4px;" alt="Image for question answering" />
    </div>
    <div style="background: #fefce8; border: 1px solid #fde68a; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
      <strong style="color: #a16207; font-size: 18px;">Question:</strong>
      <p style="font-size: 17px; line-height: 1.6; margin: 8px 0 0 0; font-weight: 500;">{{text}}</p>
      <p style="font-size: 13px; color: #6b7280; margin: 8px 0 0 0;">Question type: {{question_type}}</p>
    </div>
  </div>

allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false