Bazaar Ghost

System for indexing The Bazaar VODs from Twitch - detects matchups and extracts usernames via OCR.

Setup

Prerequisites

• Python 3.11+
• Supabase CLI
• Docker (optional)
• Streamlink, FFmpeg, Tesseract installed locally

Start Supabase

supabase start

Environment Variables

Create .env in project root:

# Required
SUPABASE_URL=http://localhost:54321
SUPABASE_SERVICE_ROLE_KEY=<from supabase start output>

# Optional
TWITCH_CLIENT_ID=<your-client-id>
TWITCH_CLIENT_SECRET=<your-client-secret>

Running SFOT Processor

Install dependencies

cd sfot
pip install -r requirements.txt

Run processor

python src/sfot.py <vod_id> [start_time] [end_time]

# Example: Process first hour of VOD 123456789
python src/sfot.py 123456789 0 3200

Run with Docker

cd sfot
docker-compose up

Configuration

Edit sfot/config.yaml:

processing:
  frame_rate: 0.2 # Frames per second to extract
  queue_size: 10
  timeout: 3200
  batch_update_interval: 20

detection:
  threshold: 0.78 # Template matching threshold
  crop_region: [271, 54, 503, 352] # w, h, x, y for nameplate
  template_path: "templates/matchup_template.png"

streamlink:
  default_stream: "480p"
  retry_attempts: 3

tesseract:
  lang: "eng"
  config: "--psm 8 --oem 3"

Storage

Detection images stored in Supabase Storage:

• Full color nameplate: {vod_id}/{timestamp}.jpg
• OCR debug frame: {vod_id}/ocr_debug/{timestamp}.jpg