Squash Thumbnail Maker
======================

A Docker-based utility for generating high-quality thumbnails from MP4 videos using intelligent frame detection.

Features:
---------
- Smart thumbnail generation using YOLOv8 pose detection to find frames with both players visible
- Fallback to simple thumbnail extraction if smart detection fails
- Process single videos or entire directories
- Configurable thumbnail quality and size
- Low priority execution to avoid system impact

Usage:
------
After installation, use the Docker container to process videos:

# Process a single video file
docker run --rm -v /path/to/videos:/data squashthumbnailmaker_image:latest /data/video.mp4

# Process all videos in a directory
docker run --rm -v /path/to/videos:/data squashthumbnailmaker_image:latest /data

# With custom settings
docker run --rm \
  -e ENABLE_SMART_DETECTION=false \
  -e THUMBNAIL_WIDTH=1280 \
  -e THUMBNAIL_QUALITY=90 \
  -v /path/to/videos:/data \
  squashthumbnailmaker_image:latest /data

Configuration Options:
----------------------
The following environment variables can be set in config/service.env or passed with -e flag:

- ENABLE_SMART_DETECTION: Use YOLO for intelligent thumbnail generation (default: true)
- FALLBACK_TIME: Seconds into video for fallback thumbnail (default: 10)
- THUMBNAIL_WIDTH: Width of generated thumbnails in pixels (default: 640)
- THUMBNAIL_QUALITY: JPEG quality 1-100 (default: 95)
- MAX_SAMPLE_TIME: Maximum seconds to sample in video (default: 180)
- SAMPLE_INTERVAL: Interval between samples in seconds (default: 1)
- NICE_LEVEL: Process priority 0-19, 19 is lowest (default: 19)

How It Works:
-------------
1. Smart Detection Mode (if enabled):
   - Samples frames throughout the video
   - Uses YOLOv8 pose model to detect players
   - Scores frames based on:
     * Number of players visible (ideally 2)
     * Face visibility (crucial for good thumbnails)
     * Player size and positioning
     * Players being centered in frame
   - Fine-tunes around the best frame for optimal result

2. Simple Mode (fallback):
   - Extracts frame at specified time (default 10 seconds)
   - Falls back to 1 second if initial extraction fails

Output:
-------
Thumbnails are saved as .jpg files in the same directory as the source video with the same base filename.
Example: video.mp4 → video.jpg

Requirements:
-------------
- Docker
- Videos should be in MP4, MOV, AVI, or MKV format
- Sufficient disk space for Docker image (~2GB with YOLO model)

Notes:
------
- The first run will download the YOLOv8 model (~6MB) if smart detection is enabled
- Processing time depends on video length and smart detection settings
- Thumbnails are only generated if they don't already exist (skip existing)