Skip to main content

Simulate YOLOv5 Segmentation

tip

This document aims to demonstrate how to use rknn-toolkit2 on an x86 PC to simulate inference of the YOLOv5 object segmentation model without the need for a development board. For the required environment setup, please refer to RKNN Installation.

Prepare the Model

In this example, we will use a pre-trained ONNX format model from the rknn_model_zoo as an example, convert the model, and perform simulated inference on the PC.

  • If you are using conda, please activate the rknn conda environment first.

    conda activate rknn
  • Download the yolov5s-seg.onnx model

    cd rknn_model_zoo/examples/yolov5_seg/model
    # Download the pre-trained yolov5s-seg.onnx model
    bash download_model.sh

    If you encounter network issues, you can visit this page to download the corresponding model into the appropriate folder.

  • Use rknn-toolkit2 to convert it into yolov5s-seg.rknn

    cd rknn_model_zoo/examples/yolov5_seg/python
    python3 convert.py ../model/yolov5s-seg.onnx <TARGET_PLATFORM>

    Parameter explanations:

    <onnx_model>: Specify the path to the ONNX model

    <TARGET_PLATFORM>: Specify the name of the NPU platform. Supported platforms can be found here

    <dtype>(optional): Specify i8 for int8 quantization or fp for fp16 quantization. The default is i8.

    <output_rknn_path>(optional): Specify the save path for the RKNN model. By default, it is saved in the same directory as the ONNX model with the filename yolov5s-seg.rknn.

Run the YOLOv5 Segmentation Simulated Inference Python Demo

  • Install the required dependencies via pip3

    pip3 install torchvision==0.11.2 pycocotools
  • Run the simulated inference program

    • Modify rknn_model_zoo/py_utils/rknn_executor.py to the following code, and make sure to back up the original code
    from rknn.api import RKNN

    class RKNN_model_container():
    def __init__(self, model_path, target=None, device_id=None) -> None:
    rknn = RKNN()
    print('--> Init runtime environment')
    if target==None:
    DATASET_PATH = '../../../datasets/COCO/coco_subset_20.txt'
    onnx_model = model_path[:-4] + 'onnx'
    print('--> Config model')
    rknn.config(mean_values=[[0, 0, 0]], std_values=[[255, 255, 255]], target_platform='rk3588')
    print('done')
    # Load model
    print('--> Loading model')
    ret = rknn.load_onnx(model=onnx_model)
    if ret != 0:
    print('Load model failed!')
    exit(ret)
    print('done')
    # Build model
    print('--> Building model')
    ret = rknn.build(do_quantization=True, dataset=DATASET_PATH)
    if ret != 0:
    print('Build model failed!')
    exit(ret)
    print('done')
    ret = rknn.init_runtime()
    else:
    ret = rknn.init_runtime(target=target, device_id=device_id)
    if ret != 0:
    print('Init runtime environment failed')
    exit(ret)
    print('done')

    self.rknn = rknn

    def run(self, inputs):
    if isinstance(inputs, list) or isinstance(inputs, tuple):
    pass
    else:
    inputs = [inputs]

    result = self.rknn.inference(inputs=inputs)
    return result
    • Modify rknn_model_zoo/examples/yolov5_seg/python/yolov5_seg.py line 260 to set the default value of target to None
    # parser.add_argument('--target', type=str, default=‘rk3566’, help='target RKNPU platform')
    parser.add_argument('--target', type=str, default=None, help='target RKNPU platform')
    • Run the simulated inference program
    python3 yolov5_seg.py --model_path ../model/yolov5s-seg.rknn --img_show
    • Simulated inference results (the simulator only simulates NPU computation results; actual effects and accuracy are based on board inference)

    yolov5_seg_result