This report outlines the progression of segmentation models used in a computer vision pipeline, moving from SegDecNet to RTMDet and eventually SegFormer. Each transition reflects growing project demands for better accuracy, faster inference, and improved handling of complex visual scenes, highlighting how modern architectures combine efficiency, scalability, and contextual understanding.
Migration from Detectron2/YOLO to RFDetR for Enhancing Precision, Latency and Architectural Efficiency
This article explains how a semantic segmentation model was used to detect waterhemp in corn fields using drone imagery. The approach identifies three categories—waterhemp, healthy corn, and stressed corn—providing farmers with detailed insights for better crop management. It covers data collection, image analysis, annotation, model results, and future improvements such as using multispectral imagery.
This article presents a curated collection of 14 free, drone-based agricultural datasets useful for training machine learning models in crop detection and segmentation. It highlights dataset sizes, classes, licenses, preprocessing steps, and augmentation methods for crops such as corn, potato, soybean, tobacco, cotton, rice, cabbage, radish, and sugar beet. The list is designed to help researchers and developers working on agricultural AI applications find suitable training data.