Grocery shelf object detection system built for NM i AI 2026.
Detects and localizes products on grocery shelves using a fine-tuned YOLOv8 model combined with SAHI (Slicing Aided Hyper Inference) for accurate detection in high-resolution images with dense, small objects.
- YOLOv8 for base detection
- SAHI for slicing large images into overlapping tiles to catch small objects
- Post-processing to merge overlapping predictions
Python · YOLOv8 · SAHI · OpenCV
Built for the Norwegian Championship in AI (NM i AI) 2026. The task involved robust product detection across shelf images with varying lighting, occlusion, and product density.