Trained YOLOX from scratch to avoid Ultralytics (iOS aircraft detect)

This project details the process of training custom YOLOX models from the ground up for aircraft detection, specifically to avoid the restrictive AGPL-3.0 license associated with Ultralytics' YOLO models. It showcases a practical solution for developers building commercial applications or side projects who require a permissive license, outlining the configuration and training steps for various YOLOX architectures like Nano, Tiny, Small, and a custom "Nanoish" variant. The author shares insights gained during the training process, including dataset preparation, configuration adjustments for model size and input resolution, and the challenges of modifying model architectures. This resource is particularly valuable for iOS developers, AR app creators, and machine learning practitioners looking for a flexible, MIT-licensed object detection solution for their projects, offering a transparent look into a real-world application of custom model training.