Skip to main content

YOLOv4 Object Detection

This document provides a detailed guide on how to use NPU hardware acceleration for inference with the YOLOv4 model on the Sirider S1.

The document is divided into two sections: Quick Start and Detailed Tutorial.

Quick Start​

Radxa provides a ready-to-use YOLOv4 object detection example that allows users to run the yolov4_tiny model with AIPU on the Sirider S1 directly. This example eliminates the need for complex model and execution code compilation, making it ideal for users who want to use the AIPU quickly without compiling from scratch. If you're interested in the full workflow, please refer to the Detailed Tutorial section.

  • Clone the repository:

    git clone https://github.com/zifeng-radxa/siriders1_NPU_yolov4_tiny_demo.git

  • Install dependencies:

    tip

    It is recommended to use virtualenv.

    cd siriders1_NPU_yolov4_tiny_demo/demo
    pip3 install -r requirements.txt

  • Run the yolov4 demo program:

    python3 yolov4_aipu.py -m [mode] -i [your_input_path] -r
    # python3 yolov4_aipu,py -m camera -r

    Parameters:

    -h, --help: Print parameter information.

    -m, --mode: Input mode selection, supports ['camera', 'video', 'image'].

    -i, --input: Input file path; please provide the file path when mode is set to ['video', 'image'].

    -r, --real_time: Real-time preview.

    -s, --save: Save output to the output folder.

    input.webp

Detailed Tutorial​

Model Conversion​

tip

This process is done on an x86 host machine. Before converting the model, install the Zhouyi SDK and complete the nn-compiler environment setup as described in the Zhouyi Z2 AIPU Usage Guide.

  • Generate quantization data:

    cd siriders1_NPU_yolov4_tiny_demo/convert
    python3 preprocess.py

  • Generate the aipu model:

    aipubuild tflite_yolo_v4_tinybuild.cfg

    The generated model is saved in ./aipu_yolov4_tiny.bin.

Compile the AIPU Executable Inference File​

Compile an executable file for inference with the Zhouyi Z2 AIPU model.

  • Copy the compile folder from the repository to the Zhouyi SDK:

    Copy the compile folder from the siriders1_NPU_yolov4_tiny_demo repository to YOUR_SDK_PATH/siengine. Be sure to replace YOUR_SDK_PATH with your actual path.

    cp -r compile YOUR_SDK_PATH/siengine

  • Cross-compilation:

    tip

    Modify the Linux_Tool_ROOT path in CMakeList.txt according to your cross-compilation toolchain path. The default is /opt/gcc-linaro-7.5.0-2019.12-x86_64_aarch64-linux-gnu/bin.

    cd YOUR_SDK_PATH/siengine/compile
    mkdir build && cd build
    cmake -DCMAKE_BUILD_TYPE=Release .. && make
    cd ..

    The output file is saved in the out folder.

  • Transfer to the board and test

    Transfer the generated aipu_yolov4_tiny.bin model file and the files in out/linux to the Sirider S1.

    Test the results using yolov4_aipu.py:

    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:linux/libs
    python3 yolov4_aipu.py -m image -i YOUR_IMAGE_PATH -r