norway1994/ainm-object-detection

NorgesGruppen Grocery Shelf Object Detection

YOLOv8x model fine-tuned for grocery product detection and classification on store shelf images. Trained for the NM i AI 2026 competition (NorgesGruppen Data task).

Results

Public leaderboard score: 0.8969

Scoring: 0.7 × detection mAP@0.5 + 0.3 × classification mAP@0.5

Model Details

• —Architecture: YOLOv8x
• —Training resolution: 1280px
• —Classes: 356 grocery product categories
• —Training data: 248 shelf images, ~22,700 annotations
• —Augmentation: copy_paste=0.3, mixup=0.15, scale=0.5, mosaic=1.0
• —Weights format: FP16 PyTorch (.pt), 132MB
• —Framework: ultralytics 8.1.0

Usage

Inference Settings

The best public score was achieved with:

• —
• —
• —(FP16)
• —Simple single-pass inference (no SAHI, no TTA, no ensemble)

GitHub

github.com/marvika/ainm-object-detection