FineDiving: Procedure-aware Action Quality Assessment

# FineDiving - Procedure-aware Action Quality Assessment
# Based on Xu, Rao, Yu, Chen, Zhou & Lu, CVPR 2022
# Dataset: https://github.com/xujinglin/FineDiving
#
# FineDiving is a fine-grained diving dataset for procedure-aware action
# quality assessment. Each dive is annotated with an action type and its
# consecutive sub-action steps (with temporal boundaries), plus the official
# judge scores, the resulting diving score, and the difficulty degree.
#
# This Potato config reproduces the annotation: mark the dive and each
# sub-action step on the timeline, record the action (dive) type and
# difficulty degree, and enter the execution and final scores.

annotation_task_name: "FineDiving - Procedure-aware Action Quality Assessment"
task_dir: "."

data_files:
  - sample-data.json

item_properties:
  id_key: "id"
  text_key: "video_url"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

annotation_schemes:
  # Step 1: mark the dive and its consecutive sub-action steps
  - name: "dive_steps"
    description: "Mark the full dive, then each consecutive sub-action step (take-off, flight, entry)."
    annotation_type: "video_annotation"
    mode: "segment"
    labels:
      - name: "dive"
        color: "#0EA5E9"
        key_value: "d"
      - name: "sub_action_step"
        color: "#22C55E"
        key_value: "s"
    frame_stepping: true
    show_timecode: true
    playback_rate_control: true
    video_fps: 25

  # Step 2: action (dive) type
  - annotation_type: text
    name: dive_type
    description: "Dive type / code (e.g. the FINA dive number and description)."

  # Step 3: difficulty degree
  - annotation_type: text
    name: difficulty_degree
    description: "Difficulty degree of the dive (as listed on the score sheet)."

  # Step 4: execution and final scores
  - annotation_type: text
    name: execution_score
    description: "Execution score (judges' raw scores or their average)."

  - annotation_type: text
    name: final_score
    description: "Final diving score (execution scaled by difficulty degree)."

annotation_instructions: |
  You will see a competitive diving clip. For each dive:

  1. Mark the temporal boundary of the whole dive on the timeline.
  2. Mark each consecutive sub-action step (for example take-off, flight, entry).
  3. Record the dive type / code and the difficulty degree.
  4. Enter the execution score and the final score.

  FineDiving pairs the semantic steps of a dive with its scores so models can
  reason about the procedure, not just the overall look of the dive.

html_layout: |
  <div style="padding: 15px; max-width: 800px; margin: auto;">
    <div style="background: #f0f9ff; border: 1px solid #bae6fd; border-radius: 8px; padding: 10px 14px; margin-bottom: 12px; font-size: 13px; color: #075985;">
      Diving clip {{clip_id}} — mark the dive steps and record the scores.
    </div>
    <video controls style="width: 100%; border-radius: 8px;" src="{{video_url}}"></video>
  </div>

allow_all_users: true
instances_per_annotator: 40
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false