Update

2020-11-06: new trained model ("0_tensor(0.5242)_lane_detection_network") (Acc: 96.81%, FP: 0.0387, FN: 0.0245, threshold = 0.36)

Dataset (TuSimple)

Pre-trained model is "804_tensor(0.5786)_lane_detection_network.pkl"

Dataset (CurveLanes)

You can download the dataset from https://github.com/xbjxh/CurveLanes. Pre-trained model is "savefile/32_tensor(1.1001)_lane_detection_network.pkl"

Parse Curvelanes

Here are several scripts for parsing and converting datasets to TuSimple style.https://github.com/pandamax/Parse_Curvelanes

ONNX Inference

• onnx_converter.py: convert trained model into onnx.
• onnx_inference.py: use onnxruntime to run converted onnx model's inference.