LightlyTrain is an all-in-one computer vision training framework covering the model development lifecycle from self-supervised pretraining of vision foundation models to fine-tuning for detection and segmentation tasks. It supports multiple model architectures including DINOv2/v3, YOLO variants, RT-DETR, and PicoDet, and includes distillation capabilities for knowledge transfer between models. The framework handles the full pipeline from unlabeled data pretraining to edge deployment optimization.