Computer Vision
hard
~45 hours
Multi-Scale Defect Detection with SAHI
Build an industrial defect detection system that handles tiny defects using SAHI (tiled inference), FPN-based detection, and a custom evaluation pipeline that tracks detection at each scale.
Skills Demonstrated
Object detection (YOLOv8/RT-DETR)
SAHI tiled inference
Multi-scale evaluation
Production deployment with TensorRT
Implementation Steps
- Collect/curate defect dataset with multi-scale annotations
- Train YOLOv8 baseline and measure small-object recall
- Implement SAHI tiled inference pipeline
- Add FPN-based detection head for tiny objects
- Build per-scale evaluation metrics (mAP@small, medium, large)
- Export to TensorRT for production-speed inference
- Create Gradio demo with live camera feed
Interview Relevance
Why this project matters for interviews
Manufacturing AI and visual inspection are massive markets. Showing you can handle the multi-scale challenge with real metrics is directly relevant to roles at Tesla, Apple Vision, and startups.