ActivityNet Captions Dense Annotation

# ActivityNet Captions Dense Annotation Configuration
# Based on Krishna et al., ICCV 2017
# Task: Segment videos and write captions for each event

annotation_task_name: "ActivityNet Dense Captioning"
task_dir: "."

data_files:
  - data.json
item_properties:
  id_key: "id"
  text_key: "video_url"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

annotation_schemes:
  - name: "event_segments"
    description: |
      Mark temporal segments for each distinct event in the video.
      Events should be semantically meaningful and non-overlapping.
    annotation_type: "video_annotation"
    mode: "segment"
    labels:
      - name: "event"
        color: "#3B82F6"
        key_value: "e"
    zoom_enabled: true
    playback_rate_control: true
    frame_stepping: true
    show_timecode: true
    timeline_height: 80

  - name: "event_caption"
    description: |
      Write a natural language description of the event you just marked.
      Be specific about WHO does WHAT. Start with a verb.
    annotation_type: text
    min_length: 10
    max_length: 200
    placeholder: "e.g., 'A man in a red shirt kicks a soccer ball into the goal'"

allow_all_users: true
instances_per_annotator: 30
annotation_per_instance: 2

annotation_instructions: |
  ## Dense Video Captioning Task

  Your goal is to segment the video into events and describe each one.

  ### Step 1: Identify Events
  - Watch the video and identify distinct events
  - Mark the START and END of each event
  - Events should be meaningful actions/happenings

  ### Step 2: Write Captions
  - Describe WHAT happens in each segment
  - Be specific: mention people, objects, actions
  - Start with a verb (e.g., "A woman picks up...")
  - Keep it concise but complete

  ### Caption Guidelines:
  - Describe visible actions, not intentions
  - Include relevant details (clothing, objects, location)
  - Use present tense
  - Don't describe audio unless relevant

  ### Example Captions:
  - "A chef chops vegetables on a cutting board"
  - "Two children run across a playground"
  - "The camera pans across a mountain landscape"