Lane Detection - LaneNet

S100 only

This sample applies only to RDK S100. The RDK S600 system image does not include the corresponding HBM model, and the related sample code is only shipped with the S100 system image; it is not supported on S600 yet.

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

Model Description

• Introduction:

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

• HBM model name: lanenet256x512.hbm

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

• Output:

  • instance_seg_logits: 3-channel map for distinguishing different lane lines

  • binary_seg_pred: binary segmentation result indicating lane region locations

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

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

Features

• Model loading

  Use hbm_runtime to load the LaneNet model and automatically parse input/output information and quantization parameters.

• Input preprocessing

  Convert the input image to RGB format, resize it to 256x512, and apply ImageNet mean and standard deviation for normalization and standardization. Finally convert to NCHW format and add a batch dimension.

• Inference execution

  Run inference with the .run() method. Outputs include instance feature maps and binary mask maps. Priority and BPU core scheduling settings are supported.

• Output postprocessing

  Reshape and normalize output tensors:

  • instance_seg_logits: output a three-channel image to visualize each lane instance

  • binary_seg_pred: output a single-channel binary map to extract lane regions

Environment Dependencies

This sample has no special environment requirements. You only need to ensure the dependencies 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	Model file path (.hbm format)	/opt/hobot/model/s100/basic/lanenet256x512.hbm
--priority	Model running priority, range 0~255; larger is higher	0
--bpu-cores	BPU core IDs used to run the model	[0]
--test-img	Test image path	/app/res/assets/input.jpg

Quick Start

• Run the model

  • Use default parameters

    python lanenet.py

  • Run with specified 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 the result

  After successful execution, results will be drawn and saved to 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.