TextVQA: Reading Text in Images for VQA

# TextVQA - Reading Text in Images
# Based on Singh et al., CVPR 2019
# Paper: https://arxiv.org/abs/1904.08920
# Dataset: https://textvqa.org/
#
# Visual question answering task requiring reading and reasoning about
# text present in natural images (signs, labels, book covers, menus, etc.).
# Annotators provide a text answer to each question and indicate answerability.
#
# Annotation Guidelines:
# 1. Examine the image carefully, paying attention to any visible text
# 2. Read the question and determine what text information is being asked about
# 3. Type your answer in the text field -- answers should be concise
# 4. Indicate whether the question is answerable from the image
# 5. If text in the image is too blurry or occluded, mark as Unanswerable

annotation_task_name: "TextVQA - Reading Text in Images"
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: Provide the answer
  - annotation_type: text
    name: answer
    description: "Type the answer to the question based on text visible in the image."
    textarea: false
    required: true
    placeholder: "Type your answer here..."

  # Step 2: Answerability judgment
  - annotation_type: radio
    name: answerability
    description: "Can this question be answered from the text visible in the image?"
    labels:
      - "Answerable"
      - "Unanswerable"
      - "Ambiguous"
    keyboard_shortcuts:
      "Answerable": "1"
      "Unanswerable": "2"
      "Ambiguous": "3"
    tooltips:
      "Answerable": "The question can be clearly answered using text visible in the image"
      "Unanswerable": "The required text is not visible, too blurry, or not present in the image"
      "Ambiguous": "The answer is unclear due to partial occlusion, multiple possible readings, or vague question"

annotation_instructions: |
  You will be shown an image along with a question that requires reading text in the image.

  For each item:
  1. Look at the image and identify any text (signs, labels, logos, numbers, etc.).
  2. Read the question carefully.
  3. Type a concise answer based on the text you can read in the image.
  4. Indicate whether the question is answerable, unanswerable, or ambiguous.

  Tips:
  - Answers should be short and precise (usually 1-3 words).
  - If the text is partially visible, provide your best reading.
  - Mark as "Unanswerable" if the relevant text is not in the image at all.

html_layout: |
  <div style="padding: 15px; max-width: 800px; margin: auto;">
    <div style="text-align: center; margin-bottom: 16px; background: #1a1a1a; padding: 12px; border-radius: 8px;">
      <img src="{{image_url}}" style="max-width: 100%; max-height: 500px; border-radius: 4px;" alt="Image for TextVQA" />
      <p style="color: #999; font-size: 12px; margin-top: 8px;">{{image_description}}</p>
    </div>
    <div style="background: #f0f9ff; border: 1px solid #bae6fd; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
      <strong style="color: #0369a1;">Question:</strong>
      <p style="font-size: 18px; line-height: 1.7; margin: 8px 0 0 0;">{{text}}</p>
    </div>
  </div>

allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 3
allow_skip: true
skip_reason_required: false