ds_task_recycling_project/docs/Task.md

1. Algorithm Choice

• Selected: YOLOv8n (lightweight version)
• Why:
  • Fast detection (0.5s/image on CPU)
  • Works well with small datasets (40 images)
  • Accurate for motherboard components

2. Hardware Impact

• Training:
  • GPU recommended (4x faster training)
  • CPU works but slower
• Deployment:
  • CPU sufficient for basic use
  • GPU better for high volume

3. Video Handling

• Approach: Process each frame individually
• Changes Needed:
  • Add frame-by-frame processing
  • Include tracking to follow memory modules
  • Optimize for speed (lower resolution helps)

Key Facts:

• Same model works for images/video
• CPU processing is practical
• No architecture changes needed between image/video modes.