YOLOv5-det Object Detection

• Enter the Radxa Model-Zoo YOLOv5 directory

  cd Radxa-Model-Zoo/sample/YOLOv5

• Download the model, choose from F32/F16/INT8 quantized models

  # F32
  wget https://github.com/radxa-edge/TPU-Edge-AI/releases/download/model-zoo/yolov5s_v6.1_3output_fp32_1b.bmodel
  
  # F16
  wget https://github.com/radxa-edge/TPU-Edge-AI/releases/download/model-zoo/yolov5s_v6.1_3output_fp16_1b.bmodel
  
  # INT8
  wget https://github.com/radxa-edge/TPU-Edge-AI/releases/download/model-zoo/yolov5s_v6.1_3output_int8_1b.bmodel

• Download test images to the data folder

  mkdir images && cd images
  wget https://github.com/radxa-edge/TPU-Edge-AI/releases/download/model-zoo/dog_bike_car.jpg

• Create a virtual environment

  It's necessary to create a virtual environment, otherwise it may affect the normal operation of other applications, please refer to here for virtual environment usage

  python3 -m virtualenv .venv
  source .venv/bin/activate

• Install python dependencies

  pip3 install --upgrade pip
  pip3 install numpy
  pip3 install https://github.com/radxa-edge/TPU-Edge-AI/releases/download/v0.1.0/sophon_arm-3.7.0-py3-none-any.whl

• Run python examples

  Under the python directory, there are a series of Python routines, as follows:

  No.	Python Routine	Description
  1	yolov5_opencv.py	Decode with OpenCV, preprocess with OpenCV, and infer with SAIL
  2	yolov5_bmcv.py	Decode with SAIL, preprocess with BMCV, and infer with SAIL

  • Run yolov5_opencv.py

    export LD_LIBRARY_PATH=/opt/sophon/libsophon-current/lib:$LD_LIBRARY_PATH
    export PYTHONPATH=$PYTHONPATH:/opt/sophon/sophon-opencv-latest/opencv-python/
    python3 python/yolov5_opencv.py --input ./images --bmodel ./yolov5s_v6.1_3output_fp16_1b.bmodel

    Parameter explanation

    yolov5_opencv.py [--input IMG_PATH] [--bmodel BMODEL]

    --input: Inference image path, you can input the entire image folder path or video path

    --bmodel: Path of the bmodel used for inference

    (.venv) linaro@bm1684:/data/ssd/docs_check/Radxa-Model-Zoo/sample/YOLOv5$ python3 python/yolov5_opencv.py --input ./images --bmodel ./yolov5s_v6.1_3output_fp16_1b.bmodel
    [BMRT][bmcpu_setup:406] INFO:cpu_lib 'libcpuop.so' is loaded.
    bmcpu init: skip cpu_user_defined
    open usercpu.so, init user_cpu_init
    [BMRT][load_bmodel:1084] INFO:Loading bmodel from [./yolov5s_v6.1_3output_fp16_1b.bmodel]. Thanks for your patience...
    [BMRT][load_bmodel:1023] INFO:pre net num: 0, load net num: 1
    INFO:root:load ./yolov5s_v6.1_3output_fp16_1b.bmodel success!
    INFO:root:1, img_file: ./images/dog_bike_car.jpg
    sampleFactor=6, cinfo->num_components=3 (1x2, 1x1, 1x1)
    Open /dev/jpu successfully, device index = 0, jpu fd = 24, vpp fd = 25
    INFO:root:result saved in ./results/yolov5s_v6.1_3output_fp16_1b.bmodel_images_opencv_python_result.json
    INFO:root:------------------ Predict Time Info ----------------------
    INFO:root:decode_time(ms): 16.56
    INFO:root:preprocess_time(ms): 31.80
    INFO:root:inference_time(ms): 20.42
    INFO:root:postprocess_time(ms): 150.16
    all done.

    The results will be saved in ./results/yolov5s_v6.1_3output_fp16_1b.bmodel_images_opencv_python_result.json

    Image results are saved in ./result/images

  • Run yolov5_bmcv.py

    export LD_LIBRARY_PATH=/opt/sophon/libsophon-current/lib:$LD_LIBRARY_PATH
    export PYTHONPATH=$PYTHONPATH:/opt/sophon/sophon-opencv-latest/opencv-python/
    python3 python/yolov5_bmcv.py  --input ./images --bmodel ./yolov5s_v6.1_3output_fp16_1b.bmodel

    Parameter explanation

    yolov5_bmcv.py [--input IMG_PATH] [--bmodel BMODEL]

    --input: Inference image path, you can input the entire image folder path or video path

    --bmodel: Path of the bmodel used for inference

    (.venv) linaro@bm1684:/data/ssd/docs_check/Radxa-Model-Zoo/sample/YOLOv5$         python3 python/yolov5_bmcv.py  --input ./images --bmodel ./yolov5s_v6.1_3output_fp16_1b.bmodel
    [BMRT][bmcpu_setup:406] INFO:cpu_lib 'libcpuop.so' is loaded.
    bmcpu init: skip cpu_user_defined
    open usercpu.so, init user_cpu_init
    [BMRT][load_bmodel:1084] INFO:Loading bmodel from [./yolov5s_v6.1_3output_fp16_1b.bmodel]. Thanks for your patience...
    [BMRT][load_bmodel:1023] INFO:pre net num: 0, load net num: 1
    INFO:root:1, img_file: ./images/dog_bike_car.jpg
    sampleFactor=6, cinfo->num_components=3 (1x2,
    
    1x1, 1x1)
    Open /dev/jpu successfully, device index = 0, jpu fd = 43, vpp fd = 44
    INFO:root:result saved in ./results/yolov5s_v6.1_3output_fp16_1b.bmodel_images_bmcv_python_result.json
    INFO:root:------------------ Predict Time Info ----------------------
    INFO:root:decode_time(ms): 25.40
    INFO:root:preprocess_time(ms): 2.68
    INFO:root:inference_time(ms): 17.16
    INFO:root:postprocess_time(ms): 153.20
    all done.

    The results will be saved in ./results/yolov5s_v6.1_3output_fp16_1b.bmodel_images_bmcv_python_result.json

    Image results are saved in ./result/images

    

    The bmcv example routine currently does not annotate class labels on images

FAQ

For more information on model conversion, model quantization details, and model accuracy testing, please refer to:

Model Conversion and Radxa Model-Zoo YOLOv5