Food Component Detection for Free Nutritious Meals

My Contribution

• Developed an object detection solution to identify food components in Free Nutritious Meals servings.
• Prepared and downloaded the YOLOv8-format detection dataset using Roboflow.
• Created training notebooks for YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, and YOLOv8x experiments.
• Compared the effect of model variants and batch sizes 8, 16, and 32 on detection performance.
• Configured and ran training experiments with up to 100 epochs using Ultralytics YOLOv8.
• Evaluated model performance using precision, recall, mAP@50, and mAP@50-95.
• Saved training configurations, per-epoch metrics, best model weights, and validation predictions for reproducible analysis.
• Analyzed validation visualizations, loss curves, metric graphs, and confusion matrices across experiments.