Lane Detection - LaneNet

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

Model Description

• Introduction:

  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 256×512, 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.

• Model Download URL (automatically downloaded by the program):

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

Functionality Description

• Model Loading

  Load the LaneNet model using hbm_runtime, which automatically parses input/output information and quantization parameters.

• Input Preprocessing

  Convert the input image to RGB format, resize it to 256×512, apply normalization and standardization using ImageNet mean and standard deviation, convert to NCHW format, and add a batch dimension.

• Inference Execution

  Perform inference using the .run() method. The outputs include the instance feature map and binary mask, with support for setting inference priority and BPU core scheduling.

• Post-processing

  Reshape and normalize the output tensors:

  • instance_seg_logits: output a 3-channel image for visualizing each lane instance.
  • binary_seg_pred: output a single-channel binary image for extracting lane regions.

Environment Dependencies

This example has no special environment requirements—just ensure the dependencies listed in pydev are installed:

pip install -r ../../requirements.txt

Directory Structure

.
├── lanenet.py                  # Main inference script
└── README.md                   # Usage instructions

Parameter Description

Parameter	Description	Default Value
--model-path	Path to the model file in .hbm format	/opt/hobot/model/s100/basic/lanenet256x512.hbm
--priority	Inference priority (range: 0–255; higher value = higher priority)	0
--bpu-cores	BPU core IDs to use for model execution	[0]
--test-img	Path to the test image	/app/res/assets/input.jpg

Quick Start

• Run the Model

  • With default parameters:

    python lanenet.py

  • With custom parameters:

    python lanenet.py \
    --model-path /opt/hobot/model/s100/basic/lanenet256x512.hbm \
    --priority 0 \
    --bpu-cores 0 \
    --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

• If the specified model path does not exist, the program will attempt to download the model automatically.