Garlic Segmentation

Project overview

Agricultural images were annotated for a computer vision dataset focused on segmenting garlic plants in field environments. This was one of the most challenging segmentation projects undertaken, owing to the plants' small size, dense vegetation, overlapping leaves, and complex agricultural backgrounds.

The goal was to create precise segmentation masks for individual garlic plants, producing high-quality training data for an AI-based segmentation model.

Work performed

• Reviewed and organised the agricultural image dataset before annotation.
• Identified individual garlic plants in field images.
• Created precise polygon segmentation masks that followed the visible boundaries of each garlic plant.
• Consistent labelling was maintained using the Garlic class.
• Partially visible plants were carefully segmented where applicable.
• Distinguished garlic plants from surrounding crops, weeds, soil, and vegetation.
• Handled overlapping leaves and closely positioned plants with detailed polygon annotations.
• Performed quality checks and corrected inaccurate, incomplete, or poorly positioned segmentation masks.
• Ensured the final annotations were suitable for training a computer vision segmentation model.

What made it hard

• Garlic plants were often very small and difficult to segment precisely.
• Thin and elongated leaves made polygon placement challenging.
• Multiple garlic plants frequently overlapped or were positioned very close to each other.
• Dense vegetation made individual plant boundaries difficult to identify.
• Some plants were partially hidden behind other plants or leaves.
• Similar-looking weeds and surrounding vegetation created additional annotation challenges.
• Complex soil and field backgrounds made object boundaries less obvious.
• Varying lighting conditions and image quality affected plant visibility.
• Maintaining precise polygon boundaries across a large dataset required significant attention to detail.

Result

A high-quality garlic segmentation dataset was completed with precise polygon annotations. The dataset was prepared for training a computer vision model capable of automatically detecting and segmenting garlic plants in agricultural fields.

Final deliverable

Accurate and detailed segmentation masks for garlic plants were delivered across the dataset. Despite the complexity of the images and the difficulty of separating small, overlapping plants, the annotations met project requirements and were ready for AI model training and agricultural computer vision applications.