hamro Computer Vision Dataset

This project utilizes YOLOv8 to detect nutritional labels on a dataset of 340 annotated images managed in Roboflow, focusing on calculating the glycemic index (GI) and glycemic load (GL) of food products. The dataset, comprising diverse label formats from various food items, is annotated with bounding boxes for the "nutritional_label" class. Preprocessing steps, including auto-orient, resize (640x640), auto-adjust contrast, dynamic crop, and filter null, ensure image quality and consistency. To enhance model robustness, image-level augmentations such as rotation (±15°), brightness (±20%), exposure (±20%), noise (1-2%), blur (3x3 kernel, 10% probability), shear (±10°), and horizontal/vertical flips (50% probability) are applied, generating approximately 1,000 images for training. The trained YOLOv8 model detects label regions, which are then processed with EasyOCR to extract key nutritional data like carbohydrates, sugars, and serving size. This data enables GI calculation using a reference database and GL computation via the formula GL = (GI × Net Carbohydrates) / 100. The system aims to assist users in understanding the glycemic impact of foods for dietary planning and health monitoring. Future enhancements may include multi-class detection of specific nutrients and mobile app integration.