环境搭建

• 本文初始环境为PyTorch 2.0.0、Python 3.8(ubuntu20.04)、Cuda 11.8

OpenMMLab基础环境

• 首先安装OpenMMLab基础环境，以下代码均在命令窗口下输入

pip install openmim
mim install mmcv-full
mim install "mmengine==0.7.2"

git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection
pip install -e . >> /dev/null

mkdir checkpoint
mkdir output
cd ..

• mim install mmcv-full这一步需要大概15~30分钟，没报错就耐心等待。

MMDeploy环境

• 再次声明，下面代码均在命令窗中输入运行！！

git clone https://github.com/open-mmlab/mmdeploy.git --recursive
cd mmdeploy

# 备份镜像源
sudo cp /etc/apt/sources.list /etc/apt/sources.list.backup
# 更换镜像源
sudo sed -i "s@http://.*archive.ubuntu.com@https://mirrors.tuna.tsinghua.edu.cn@g" /etc/apt/sources.list
sudo sed -i "s@http://.*security.ubuntu.com@https://mirrors.tuna.tsinghua.edu.cn@g" /etc/apt/sources.list
sudo apt-get update

python tools/scripts/build_ubuntu_x64_ort.py

• 更改环境变量，依然是在命令行输入

export PYTHONPATH=$(pwd)/build/lib:$PYTHONPATH
export LD_LIBRARY_PATH=$(pwd)/build/lib:$(pwd)/../mmdeploy-dep/onnxruntime-linux-x64-1.8.1/lib:$LD_LIBRARY_PATH

• 可以使用脚本检查环境安装状态python tools/check_env.py

07/14 19:14:16 - mmengine - INFO - **********Environmental information**********
07/14 19:14:17 - mmengine - INFO - sys.platform: linux
07/14 19:14:17 - mmengine - INFO - Python: 3.8.10 (default, Jun  4 2021, 15:09:15) [GCC 7.5.0]
07/14 19:14:17 - mmengine - INFO - CUDA available: True
07/14 19:14:17 - mmengine - INFO - numpy_random_seed: 2147483648
07/14 19:14:17 - mmengine - INFO - GPU 0: NVIDIA GeForce RTX 2080 Ti
07/14 19:14:17 - mmengine - INFO - CUDA_HOME: /usr/local/cuda
07/14 19:14:17 - mmengine - INFO - NVCC: Cuda compilation tools, release 11.8, V11.8.89
07/14 19:14:17 - mmengine - INFO - GCC: gcc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
07/14 19:14:17 - mmengine - INFO - PyTorch: 2.0.0+cu118
07/14 19:14:17 - mmengine - INFO - PyTorch compiling details: PyTorch built with:
  - GCC 9.3
  - C++ Version: 201703
  - Intel(R) oneAPI Math Kernel Library Version 2022.2-Product Build 20220804 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v2.7.3 (Git Hash 6dbeffbae1f23cbbeae17adb7b5b13f1f37c080e)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 11.8
  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90
  - CuDNN 8.7
  - Magma 2.6.1
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.8, CUDNN_VERSION=8.7.0, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_DISABLE_GPU_ASSERTS=ON, TORCH_VERSION=2.0.0, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, 

07/14 19:14:17 - mmengine - INFO - TorchVision: 0.15.1+cu118
07/14 19:14:17 - mmengine - INFO - OpenCV: 4.8.0
07/14 19:14:17 - mmengine - INFO - MMEngine: 0.7.2
07/14 19:14:17 - mmengine - INFO - MMCV: 2.0.0
07/14 19:14:17 - mmengine - INFO - MMCV Compiler: GCC 9.3
07/14 19:14:17 - mmengine - INFO - MMCV CUDA Compiler: 11.8
07/14 19:14:17 - mmengine - INFO - MMDeploy: 1.2.0+0a8cbe2
07/14 19:14:17 - mmengine - INFO - 

07/14 19:14:17 - mmengine - INFO - **********Backend information**********
07/14 19:14:17 - mmengine - INFO - tensorrt:    None
07/14 19:14:17 - mmengine - INFO - ONNXRuntime: 1.8.1
07/14 19:14:17 - mmengine - INFO - ONNXRuntime-gpu:     None
07/14 19:14:17 - mmengine - INFO - ONNXRuntime custom ops:      Available
07/14 19:14:17 - mmengine - INFO - pplnn:       None
07/14 19:14:17 - mmengine - INFO - ncnn:        None
07/14 19:14:17 - mmengine - INFO - snpe:        None
07/14 19:14:17 - mmengine - INFO - openvino:    None
07/14 19:14:17 - mmengine - INFO - torchscript: 2.0.0+cu118
07/14 19:14:17 - mmengine - INFO - torchscript custom ops:      NotAvailable
07/14 19:14:18 - mmengine - INFO - rknn-toolkit:        None
07/14 19:14:18 - mmengine - INFO - rknn-toolkit2:       None
07/14 19:14:18 - mmengine - INFO - ascend:      None
07/14 19:14:18 - mmengine - INFO - coreml:      None
07/14 19:14:18 - mmengine - INFO - tvm: None
07/14 19:14:18 - mmengine - INFO - vacc:        None
07/14 19:14:18 - mmengine - INFO - 

07/14 19:14:18 - mmengine - INFO - **********Codebase information**********
07/14 19:14:18 - mmengine - INFO - mmdet:       3.1.0
07/14 19:14:18 - mmengine - INFO - mmseg:       None
07/14 19:14:18 - mmengine - INFO - mmpretrain:  None
07/14 19:14:18 - mmengine - INFO - mmocr:       None
07/14 19:14:18 - mmengine - INFO - mmagic:      None
07/14 19:14:18 - mmengine - INFO - mmdet3d:     None
07/14 19:14:18 - mmengine - INFO - mmpose:      None
07/14 19:14:18 - mmengine - INFO - mmrotate:    None
07/14 19:14:18 - mmengine - INFO - mmaction:    None
07/14 19:14:18 - mmengine - INFO - mmrazor:     None
07/14 19:14:18 - mmengine - INFO - mmyolo:      None

YOLO-x转ONNX格式

• 下载YOLO-x模型的预训练权重

cd ..
cd mmdetection

# 下载YOLOX-X权重
!weget https://download.openmmlab.com/mmdetection/v2.0/yolox/yolox_x_8x8_300e_coco/yolox_x_8x8_300e_coco_20211126_140254-1ef88d67.pth -P checkpoint
cd ..

• 使用转换脚本将YOLO-x模型转换为ONNX格式

python mmdeploy/tools/deploy.py \
    mmdeploy/configs/mmdet/detection/detection_onnxruntime_dynamic.py \
    mmdetection/configs/yolox/yolox_x_8xb8-300e_coco.py \
    mmdetection/checkpoint/yolox_x_8x8_300e_coco_20211126_140254-1ef88d67.pth \
    mmdetection/demo/demo.jpg \
    --work-dir mmdeploy_models/mmdet/yolox \
    --device cpu \
    --show \
    --dump-info

• 在经过上述代码后，工作目录下的文件组织信息为：

 - mmdeploy_models
     - mmdet
         - yolox
        	 - deploy.json
        	 - detail.json
        	 - end2end.onnx
        	 - pipeline.json

• 其中，end2end.onnx: 推理引擎文件。可用ONNX Runtime推理。*.json: mmdeploy SDK推理所需的meta 信息。

模型推理

后端推理

from mmdeploy.apis.utils import build_task_processor
from mmdeploy.utils import get_input_shape, load_config
import torch

deploy_cfg = '/kaggle/working/mmdeploy/configs/mmdet/detection/detection_onnxruntime_dynamic.py'
model_cfg = '/kaggle/working/mmdetection/configs/yolox/yolox_x_8xb8-300e_coco.py'
device = 'cpu'
backend_model = ['/kaggle/input/yolo-x-onnx-model/end2end.onnx']
image = '/kaggle/working/mmdetection/demo/demo.jpg'

deploy_cfg, model_cfg = load_config(deploy_cfg, model_cfg)

task_processor = build_task_processor(model_cfg, deploy_cfg, device)
model = task_processor.build_backend_model(backend_model)

input_shape = get_input_shape(deploy_cfg)
model_inputs, _ = task_processor.create_input(image, input_shape)

with torch.no_grad():
    result = model.test_step(model_inputs)

task_processor.visualize(
    image=image,
    model=model,
    result=result[0],
    window_name='visualize',
    output_file='output_detection.png')

SDK推理

from mmdeploy_runtime import Detector
import cv2

img = cv2.imread('/kaggle/working/mmdetection/demo/demo.jpg')

# create a detector
detector = Detector(model_path='/kaggle/input/yolo-x-onnx-model', device_name='cpu', device_id=0)
# perform inference
bboxes, labels, masks = detector(img)

# visualize inference result
indices = [i for i in range(len(bboxes))]
for index, bbox, label_id in zip(indices, bboxes, labels):
  [left, top, right, bottom], score = bbox[0:4].astype(int), bbox[4]
  if score < 0.3:
    continue

  cv2.rectangle(img, (left, top), (right, bottom), (0, 255, 0))

cv2.imwrite('output_detection.png', img)