RefCOCO - Referring Expression Grounding

# RefCOCO - Referring Expression Grounding
# Based on Yu et al., ECCV 2016
# Paper: https://arxiv.org/abs/1608.00272
# Dataset: https://github.com/lichengunc/refer
#
# Visual grounding task: given an image and a referring expression, annotators
# draw a bounding box around the object described by the expression. This tests
# the ability to link natural language to spatial image regions.
#
# Annotation Guidelines:
# 1. Read the referring expression carefully
# 2. Examine the image and identify the object being described
# 3. Draw a tight bounding box around the referred object
# 4. If the expression is ambiguous, optionally write a clarification
# 5. The bounding box should be as tight as possible around the object

annotation_task_name: "RefCOCO - Referring Expression Grounding"
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: Draw bounding box around the referred object
  - annotation_type: image_annotation
    name: referred_object
    description: "Draw a bounding box around the object described by the referring expression."
    tools:
      - bbox
    labels:
      - "Referred Object"

  # Step 2: Optionally provide a clarifying expression
  - annotation_type: text
    name: clarification
    description: "If the original expression is ambiguous, write a more precise referring expression."
    textarea: false
    required: false
    placeholder: "Optional: write a clearer referring expression..."

annotation_instructions: |
  You will see an image and a referring expression that describes one specific object in the image.

  For each item:
  1. Read the referring expression below the image.
  2. Identify which object in the image is being described.
  3. Use the bounding box tool to draw a tight box around that object.
  4. If the expression is ambiguous or could refer to multiple objects,
     draw the box around the most likely referent and optionally write
     a clearer expression in the text field.

  Tips:
  - Draw the box as tightly as possible around the entire object.
  - Referring expressions often use spatial relationships (e.g., "the dog on the left").
  - Consider attributes like color, size, and position to disambiguate.

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 grounding" />
    </div>
    <div style="background: #fef3c7; border: 1px solid #fbbf24; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
      <strong style="color: #92400e;">Referring Expression:</strong>
      <p style="font-size: 18px; line-height: 1.7; margin: 8px 0 0 0; font-style: italic;">{{text}}</p>
    </div>
  </div>

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