Unofficial_implemention_of_lanenet_model

# jetson To run LaneNet on Jetson (e.g. Xavier NX, Nano), a custom Docker container is needed. ## Docker Usage ### Build ```bash docker build -t lanenet-jetson:latest . ``` ### Run Allow external applications to connect to the host’s X display: ```bash xhost + ``` Start an interactive session in the container: ```bash docker run -it --rm \ --runtime nvidia \ --network host \ -e DISPLAY=$DISPLAY \ -v /tmp/.X11-unix/:/tmp/.X11-unix \ -v $PWD/..:/code \ lanenet-jetson:latest ``` ## Using TensorRT for inference ### Converting TensorFlow model to TensorRT Freeze meta & checkpoint files: ```bash python tensorrt/freeze_graph.py --weights_path model/tusimple_lanenet/tusimple_lanenet.ckpt --save_path model/lanenet.pb ``` ### Converting frozen graph to ONNX ```bash python -m tf2onnx.convert \ --input ./model/lanenet.pb \ --output ./model/lanenet.onnx \ --inputs lanenet/input_tensor:0 \ --outputs lanenet/final_binary_output:0,lanenet/final_pixel_embedding_output:0 ``` ### Running inference with a video source ```bash python3 tensorrt/trt_inference.py \ --onnx_file ./model/lanenet.onnx \ --video_src ./data/tusimple_test_video/0.mp4 \ --engine_file ./tensorrt/jetson.engine ```