YOLOv5推理部署

# 使用OpenCV 4.5.4推理YOLOv5模型，分别使用C++，Python和ROS实现。

基于yolov5-opencv-cpp-python (opens new window)修改。 在原代码的基础上使用CMake编译（能够更方便的定义程序路径），并且加入了对ROS传入图片的支持。

代码地址：github.com/YellowAndGreen/Yolov5-OpenCV-Cpp-Python-ROS

# 代码解释

变量net_path定义了onnx网络模型路径.
变量 class_path定义了分类文件的路径.
ROS节点订阅**/image话题来获取输入图像. NOTE: The code depends on the output dimension of your network model, which means the variables dimensions and rows in ros.cpp should be the exact same size of output dimensions. 注意: 网络推理的计算取决于网络模型的输出维度, 也就是说ros.cpp中的变量dimensions and rows**应该与其一致.

# 环境配置

任何Linux OS (在Ubuntu 18.04上测试)
OpenCV 4.5.4+
Python 3.7+(可选)
GCC 9.0+(可选)
ROS melodic(可选)

注意!!! 先于4.5.4的OpenCV版本不会正常运行。

# 使用ROS/C++推理

C++/ROS代码在Yolo_ROS/ros.cpp (opens new window)。

git clone https://github.com/YellowAndGreen/Yolov5-OpenCV-Cpp-Python-ROS.git
cd Yolov5-OpenCV-Cpp-Python-ROS/Yolo_ROS
mkdir build && cd build
cmake ../
make
./yolo_ros

1
2
3
4
5
6

# 使用python推理

Python代码在python/yolo.py (opens new window).

git clone https://github.com/YellowAndGreen/Yolov5-OpenCV-Cpp-Python-ROS.git
cd Yolov5-OpenCV-Cpp-Python-ROS
python python/yolo.py

1
2
3

使用GPU运行：

git clone https://github.com/YellowAndGreen/Yolov5-OpenCV-Cpp-Python-ROS.git
cd Yolov5-OpenCV-Cpp-Python-ROS
python python/yolo.py cuda
python python/yolo-tiny.py cuda

1
2
3
4

# 使用C++推理

C++代码在cpp/yolo.cpp (opens new window).

git clone https://github.com/YellowAndGreen/Yolov5-OpenCV-Cpp-Python-ROS.git
cd Yolov5-OpenCV-Cpp-Python-ROS/cpp
mkdir build && cd build
cmake ../
make
./yolo_example

1
2
3
4
5
6

# 导出Yolov5 模型到onnx格式

https://github.com/ultralytics/yolov5/issues/251

我的指令是:

git clone https://github.com/ultralytics/yolov5
cd yolov5
pip install -r requirements.txt

1
2
3

然后转换模型:

$ python3 export.py --weights yolov5n.pt --img 640 --include onnx
export: data=data/coco128.yaml, weights=['yolov5n.pt'], imgsz=[640], batch_size=1, device=cpu, half=False, inplace=False, train=False, optimize=False, int8=False, dynamic=False, simplify=False, opset=12, verbose=False, workspace=4, nms=False, agnostic_nms=False, topk_per_class=100, topk_all=100, iou_thres=0.45, conf_thres=0.25, include=['onnx']
YOLOv5   v6.0-192-g436ffc4 torch 1.10.1+cu102 CPU

Fusing layers... 
Model Summary: 213 layers, 1867405 parameters, 0 gradients

PyTorch: starting from yolov5n.pt (4.0 MB)

ONNX: starting export with onnx 1.10.2...
/home/user/workspace/smartcam/yolov5/models/yolo.py:57: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if self.onnx_dynamic or self.grid[i].shape[2:4] != x[i].shape[2:4]:
ONNX: export success, saved as yolov5n.onnx (7.9 MB)

Export complete (1.33s)
Results saved to /home/doleron/workspace/smartcam/yolov5
Visualize with https://netron.app
Detect with `python detect.py --weights yolov5n.onnx` or `model = torch.hub.load('ultralytics/yolov5', 'custom', 'yolov5n.onnx')
Validate with `python val.py --weights yolov5n.onnx`
$

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

编辑

上次更新: 2023/12/12, 21:53:22