window 安装onnx和onnxruntime-gpu（python版）

目录

1. conda_py37.yaml文件

2. 命令行安装

3. 效果 

4. 代码测试环境

---

配置环境cuda10.2，onnx1.8.0，onnxruntime-gpu=1.6.0 

其他版本对应关系：

CUDA - onnxruntimeInstructions to execute ONNX Runtime applications with CUDAhttps://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html 

1. conda_py37.yaml文件

name: test_onnx
channels:
  - Esri
  - conda-forge
  - defaults
dependencies:
  - _pytorch_select=1.2.0=gpu
  - blas=1.0=mkl
  - ca-certificates=2021.5.25=haa95532_1
  - certifi=2021.5.30=py37haa95532_0
  - cffi=1.14.5=py37hd8e9650_0
  - cudatoolkit=10.2.89=h74a9793_1
  - intel-openmp=2019.4=245
  - libblas=3.8.0=14_mkl
  - libcblas=3.8.0=14_mkl
  - liblapack=3.8.0=14_mkl
  - libmklml=2019.0.5=haa95532_0
  - libprotobuf=3.14.0=h7755175_0
  - mkl=2019.4=245
  - mkl-service=2.3.0=py37hfa6e2cd_0
  - ninja=1.10.2=h5362a0b_0
  - numpy=1.20.3=py37hcbcd69c_1
  - onnx=1.8.0=py37hb2e8ee1_1
  - openssl=1.1.1k=h2bbff1b_0
  - pip=21.1.2=pyhd8ed1ab_0
  - protobuf=3.14.0=py37hf2a7229_1
  - pycparser=2.20=pyh9f0ad1d_2
  - python=3.7.10=0
  - python_abi=3.7=1_cp37m
  - setuptools=49.6.0=py37h03978a9_3
  - six=1.15.0=py_0
  - sqlite=3.35.5=h8ffe710_0
  - tbb=2021.2.0=h2d74725_0
  - typing-extensions=3.10.0.0=hd8ed1ab_0
  - typing_extensions=3.10.0.0=pyha770c72_0
  - vc=14.1=h869be7e_1
  - vs2015_runtime=14.16.27012=hf0eaf9b_0
  - wheel=0.36.2=pyhd3deb0d_0
  - wincertstore=0.2=py37h03978a9_1006
  - zlib=1.2.11=h62dcd97_1010
  - pip:
    - altgraph==0.17
    - flatbuffers==2.0
    - future==0.18.2
    - importlib-metadata==4.5.0
    - onnxruntime-gpu==1.6.0
    - opencv-contrib-python==4.5.2.52
    - opencv-python==4.5.2.52
    - pefile==2021.5.24
    - pyinstaller==4.3
    - pyinstaller-hooks-contrib==2021.1
    - pylibdmtx==0.1.9
    - pywin32-ctypes==0.2.0
    - zipp==3.4.1
prefix: D:\ProgramData\Anaconda3\envs\test_onnx

2. 命令行安装

# 安装
conda env create -f conda_py37.yaml
# 激活环境
conda activate test_onnx
# 确认环境
conda list

3. 效果 

4. 代码测试环境

下载onnx权重进行测试

alexnet.onnx权重-互联网文档类资源-CSDN下载生成该权重代码：importtorchimporttorchvision#useTr更多下载资源、学习资料请访问CSDN下载频道.https://download.csdn.net/download/jizhidexiaoming/80233035

import time

import onnx
import numpy as np
import onnxruntime as ort


# load onnx model
model = onnx.load_model('alexnet.onnx')
onnx.checker.check_model(model=model)
onnx.helper.printable_graph(graph=model.graph)


# onnx_runtime val
# sess_options = ort.SessionOptions()
# sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_DISABLE_ALL
# sess_options.optimized_model_filepath = "optimized_mnist_original.onnx"
# session = ort.InferenceSession("mnist_original.onnx", sess_options)
ort_session = ort.InferenceSession("alexnet.onnx")
while 1:
    start = time.time()
    outputs = ort_session.run(None, {'inputs': np.random.randn(10, 3, 224, 224).astype(np.float32)})  # inputs is same to orig

    print(time.time() - start)