Lane Detection - LaneNet

This example runs the LaneNet model on the BPU to perform instance segmentation and binary segmentation of lane lines, and saves the resulting images locally. The example code is located in the /app/cdev_demo/bpu/06_lane_detection_sample/01_lanenet/ directory.

Model Description

• Overview:

  LaneNet is a semantic segmentation model designed for real-time lane detection. It employs normalization and standardization during image preprocessing, making it suitable for road scene analysis in autonomous driving and ADAS systems. This example uses the quantized model lanenet256x512.hbm, which supports BPU inference acceleration.

• HBM Model Name: lanenet256x512.hbm

• Input Format: RGB, sized 256x512, normalized to [0,1] and then standardized

• Outputs:

  • instance_seg_logits: a 3-channel map used to distinguish different lane instances

  • binary_seg_pred: binary segmentation result indicating lane regions

Functionality Description

• Model Loading

  Loads the LaneNet model and automatically parses partial model metadata.

• Input Preprocessing

  Converts the input image to RGB format, resizes it to 256x512, applies normalization and standardization using ImageNet mean and standard deviation, converts it to NCHW format, and adds a batch dimension.

• Inference Execution

  Performs inference using the .infer() method, producing an instance feature map and a binary mask.

• Post-processing

  Reshapes and normalizes the output tensors:

  • instance_seg_logits: outputs a three-channel image for visualizing individual lane instances

  • binary_seg_pred: outputs a single-channel binary image for extracting lane regions

Environment Dependencies

Before compiling and running, ensure the following dependencies are installed:

sudo apt update
sudo apt install libgflags-dev

Directory Structure

.
|-- CMakeLists.txt            # CMake build script: target/dependencies/include and link configuration
|-- README.md                 # Usage instructions (this file)
|-- inc
|   `-- lanenet.hpp           # LaneNet inference wrapper header: interfaces for loading/preprocessing/inference/post-processing
`-- src
    |-- lanenet.cc            # LaneNet inference implementation: pre/post-processing and inference calls
    `-- main.cc               # Program entry point: parse arguments → full pipeline → save instance/binary results

Build the Project

• Configuration and Compilation

  mkdir build && cd build
  cmake ..
  make -j$(nproc)

Model Download

If the model is not found at runtime, download it using the following command:

wget https://archive.d-robotics.cc/downloads/rdk_model_zoo/rdk_s100/Lanenet/lanenet256x512.hbm

Parameter Description

Parameter	Description	Default Value
--model_path	Path to the model file (.hbm format)	/opt/hobot/model/s100/basic/lanenet256x512.hbm
--test_img	Path to the input test image	/app/res/assets/input.jpg

Quick Start

• Run the Model

  • Ensure you are in the build directory
  • Run with default parameters

    ./lanenet

  • Run with specified parameters

    ./lanenet \
        --model_path /opt/hobot/model/s100/basic/lanenet256x512.hbm \
        --test_img   /app/res/assets/input.jpg

• View Results

  Upon successful execution, the results will be saved as instance_pred.png and binary_pred.png:

  Results saved to: instance_pred.png and binary_pred.png

Notes

• Output results are saved as instance_pred.png and binary_pred.png; users can inspect them directly.

• For more information on deployment options or model support, please refer to the official documentation or contact platform technical support.