README.md 3.48 KB
Newer Older
Your Name's avatar
Your Name committed
1
2
3
4
5
6
7
8
# YoloV5

## 模型介绍

YoloV5是一种单阶段目标检测算法,该算法在YOLOV4的基础上添加了一些新的改进思路,使其速度与精度都得到了极大的性能提升。

## 模型结构

Your Name's avatar
Your Name committed
9
YoloV5模型的主要改进思路有以下几点:
Your Name's avatar
Your Name committed
10
11
12
13
14
15

- 输入端的Mosaic数据增强、自适应锚框计算、自适应图像缩放操作;
- 主干网络的Focus结构与CSP结构;
- Neck端的FPN+PAN结构;
- 输出端的损失函数GIOU_Loss以及预测框筛选的DIOU_nms。

shizhm's avatar
shizhm committed
16
## Python版本推理
liucong's avatar
liucong committed
17

shizhm's avatar
shizhm committed
18
下面介绍如何运行Python代码示例,Python示例的详细说明见Doc目录下的Tutorial_Python.md。
liucong's avatar
liucong committed
19

shizhm's avatar
shizhm committed
20
### 下载镜像
Your Name's avatar
Your Name committed
21
22

在光源可拉取推理的docker镜像,YoloV5工程推荐的镜像如下:
Your Name's avatar
Your Name committed
23

Your Name's avatar
Your Name committed
24
25
26
```python
docker pull image.sourcefind.cn:5000/dcu/admin/base/custom:ort1.14.0_migraphx3.0.0-dtk22.10.1
```
Your Name's avatar
Your Name committed
27

shizhm's avatar
shizhm committed
28
### 设置Python环境变量
liucong's avatar
liucong committed
29

shizhm's avatar
shizhm committed
30
31
32
33
34
```
export PYTHONPATH=/opt/dtk/lib:$PYTHONPATH
```

### 安装依赖
liucong's avatar
liucong committed
35
36
37

```
# 进入python示例目录
shizhm's avatar
shizhm committed
38
cd <path_to_yolov5_migraphx>/Python
liucong's avatar
liucong committed
39
40
41

# 安装依赖
pip install -r requirements.txt
shizhm's avatar
shizhm committed
42
```
liucong's avatar
liucong committed
43

shizhm's avatar
shizhm committed
44
### 运行示例
shizhm's avatar
shizhm committed
45

46
YoloV5模型的推理示例程序是YoloV5_infer_migraphx.py,本示例执行YOLOV5动态shape推理,在Python目录下使用如下命令运行该推理示例:
shizhm's avatar
shizhm committed
47
48

```
49
50
51
52
# 开启环境变量
export MIGRAPHX_DYNAMIC_SHAPE=1

# 运行示例
liucong's avatar
liucong committed
53
54
55
56
57
58
59
60
61
62
python YoloV5_infer_migraphx.py \
	--imgpath 测试图像路径 \ 
	--modelpath onnx模型路径 \
	--objectThreshold 判断是否有物体阈值,默认0.5 \
	--confThreshold 置信度阈值,默认0.25 \
	--nmsThreshold nms阈值,默认0.5 \
```

程序运行结束会在当前目录生成YoloV5检测结果图像。

63
64
65
<img src="./Resource/Images/Result0.jpg" alt="Result_2" style="zoom: 50%;" />

<img src="./Resource/Images/Result1.jpg" alt="Result1" style="zoom: 50%;" />
liucong's avatar
liucong committed
66
67
68

## C++版本推理

shizhm's avatar
shizhm committed
69
下面介绍如何运行C++代码示例,C++示例的详细说明见Doc目录下的Tutorial_Cpp.md。
liucong's avatar
liucong committed
70

shizhm's avatar
shizhm committed
71
72
73
74
75
76
77
### 下载镜像

在光源中下载MIGraphX镜像: 

```
docker pull image.sourcefind.cn:5000/dcu/admin/base/custom:ort1.14.0_migraphx3.0.0-dtk22.10.1
```
liucong's avatar
liucong committed
78

Your Name's avatar
Your Name committed
79
### 安装Opencv依赖
Your Name's avatar
Your Name committed
80

Your Name's avatar
Your Name committed
81
82
83
```python
cd <path_to_migraphx_samples>
sh ./3rdParty/InstallOpenCVDependences.sh
Your Name's avatar
Your Name committed
84
```
Your Name's avatar
Your Name committed
85
86
87
88
89
90
91
92
93
94
95

### 修改CMakeLists.txt

- 如果使用ubuntu系统,需要修改CMakeLists.txt中依赖库路径:
  将"${CMAKE_CURRENT_SOURCE_DIR}/depend/lib64/"修改为"${CMAKE_CURRENT_SOURCE_DIR}/depend/lib/"

- **MIGraphX2.3.0及以上版本需要c++17**


### 安装OpenCV并构建工程

Your Name's avatar
Your Name committed
96
```
Your Name's avatar
Your Name committed
97
98
99
100
rbuild build -d depend
```

### 设置环境变量
Your Name's avatar
Your Name committed
101

Your Name's avatar
Your Name committed
102
103
104
将依赖库依赖加入环境变量LD_LIBRARY_PATH,在~/.bashrc中添加如下语句:

**Centos**:
Your Name's avatar
Your Name committed
105
106

```
shizhm's avatar
shizhm committed
107
export LD_LIBRARY_PATH=<path_to_yolov5_migraphx>/depend/lib64/:$LD_LIBRARY_PATH
Your Name's avatar
Your Name committed
108
109
110
111
112
```

**Ubuntu**:

```
shizhm's avatar
shizhm committed
113
export LD_LIBRARY_PATH=<path_to_yolov5_migraphx>/depend/lib/:$LD_LIBRARY_PATH
Your Name's avatar
Your Name committed
114
115
```

Your Name's avatar
Your Name committed
116
117
118
119
120
121
然后执行:

```
source ~/.bashrc
```

shizhm's avatar
shizhm committed
122
### 运行示例
Your Name's avatar
Your Name committed
123

124
成功编译YoloV5工程后,执行如下命令运行动态shape推理该示例:
Your Name's avatar
Your Name committed
125
126

```
shizhm's avatar
shizhm committed
127
# 进入yolov5 migraphx工程根目录
shizhm's avatar
shizhm committed
128
cd <path_to_yolov5_migraphx> 
Your Name's avatar
Your Name committed
129

liucong's avatar
liucong committed
130
131
# 进入build目录
cd ./build/
Your Name's avatar
Your Name committed
132

133
134
135
# 开启环境变量
export MIGRAPHX_DYNAMIC_SHAPE=1

liucong's avatar
liucong committed
136
137
# 执行示例程序
./YOLOV5
Your Name's avatar
Your Name committed
138
```
Your Name's avatar
Your Name committed
139

140
141
142
程序运行结束会在build目录生成YoloV5动态shape推理检测结果图像。

<img src="./Resource/Images/Result0.jpg" alt="Result" style="zoom:50%;" />
Your Name's avatar
Your Name committed
143

144
<img src="./Resource/Images/Result1.jpg" alt="Result" style="zoom:50%;" />
Your Name's avatar
Your Name committed
145

shizhm's avatar
shizhm committed
146
## 源码仓库及问题反馈
Your Name's avatar
Your Name committed
147
148
149
150
151
152

​		https://developer.hpccube.com/codes/modelzoo/yolov5_migraphx

## 参考

​		https://github.com/ultralytics/yolov5