hatlem/nmiai-norgesgruppen

NM i AI 2026 — NorgesGruppen Object Detection Models

YOLO26-x multi-class object detection models trained for the NorgesGruppen grocery shelf detection task.

Ensemble (submission #8: mAP@0.5 = 0.9139)

• ensemble/pseudo_best.onnx — Pseudo-label model (mAP50=0.789)
• ensemble/fold0_best.onnx — K-fold 0 (mAP50=0.726)
• ensemble/fold2_best.onnx — K-fold 2 (mAP50=0.749)

Combined with WBF (Weighted Box Fusion) + TTA.

Additional Models

• folds/ — K-fold 1, 4
• multi/ — Multi-class v2
• single/ — Full-data YOLO26 single model
• progressive/ — Progressive resize pretrained
• pseudo/ — Pseudo-label long training
• aug/ — Augmentation v2
• highres/ — 1600px resolution

License

MIT