Complete PyTorch/Lightning computer vision pipeline for image classification and object detection. Use when users need to (1) download and preprocess image datasets, (2) train deep learning models (ResNet, EfficientNet, ViT, Swin Transformer, ConvNeXt, etc.) with easy model experimentation, (3) set up training environments (local GPU with CUDA/Apple M1-M4, AWS, GCP, Colab), (4) track experiments with WandB, or (5) evaluate and optimize vision models. Includes both vanilla PyTorch and PyTorch Lightning implementations, 25+ model architectures, optimized Apple Silicon support for M3/M4, and document understanding models (DiT, LayoutLMv3). Supports full workflow from data collection to model deployment with clean, production-ready code.