PytorchAutoDrive is a pure Python framework implementing semantic segmentation models (ERFNet, ENet, DeepLab, FCN) and lane detection models (SCNN, RESA, LSTR, LaneATT, BézierLaneNet) for autonomous driving. It supports full-stack development from research training and fair benchmarking via config-based implementations to deployment via ONNX and TensorRT. The framework supports datasets including Cityscapes, PASCAL VOC, CULane, and TuSimple, with mixed precision training and tensorboard logging.