README.md 2.93 KB
Newer Older
Your Name's avatar
Your Name committed
1
# YoloV3
shizhm's avatar
shizhm committed
2

Your Name's avatar
Your Name committed
3
4
5
6
7
8
9
10
## 模型介绍

YOLOV3是由Joseph Redmon和Ali Farhadi在2018年提出的单阶段目标检测模型,主要用于自然场景的目标检测。

## 模型结构

算法基本思想首先通过特征提取网络对输入提取特征,backbone部分由YOLOV2时期的Darknet19进化至Darknet53加深了网络层数,引入了Resnet中的跨层加和操作;然后结合不同卷积层的特征实现多尺度训练,一共有13x13、26x26、52x52三种分辨率,分别用来预测大、中、小的物体;每种分辨率的特征图将输入图像分成不同数量的格子,每个格子预测B个bounding box,每个bounding box预测内容包括: Location(x, y, w, h)、Confidence Score和C个类别的概率,因此YOLOv3输出层的channel数为B*(5 + C)。YOLOv3的loss函数也有三部分组成:Location误差,Confidence误差和分类误差。参考论文地址:https://arxiv.org/abs/1804.02767

Your Name's avatar
Your Name committed
11
12
13
## 构建安装

在光源可拉取推理的docker镜像,YoloV3工程推荐的镜像如下:
Your Name's avatar
Your Name committed
14

Your Name's avatar
Your Name committed
15
16
17
```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
18

Your Name's avatar
Your Name committed
19
### 安装Opencv依赖
Your Name's avatar
Your Name committed
20

Your Name's avatar
Your Name committed
21
22
23
```python
cd <path_to_migraphx_samples>
sh ./3rdParty/InstallOpenCVDependences.sh
Your Name's avatar
Your Name committed
24
```
Your Name's avatar
Your Name committed
25
26
27
28
29
30
31
32
33
34
35

### 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
36
```
Your Name's avatar
Your Name committed
37
38
39
40
rbuild build -d depend
```

### 设置环境变量
Your Name's avatar
Your Name committed
41

Your Name's avatar
Your Name committed
42
43
44
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
45
46

```
Your Name's avatar
Your Name committed
47
48
49
50
51
52
53
export LD_LIBRARY_PATH=<path_to_migraphx_samples>/depend/lib64/:$LD_LIBRARY_PATH
```

**Ubuntu**:

```
export LD_LIBRARY_PATH=<path_to_migraphx_samples>/depend/lib/:$LD_LIBRARY_PATH
Your Name's avatar
Your Name committed
54
55
```

Your Name's avatar
Your Name committed
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
然后执行:

```
source ~/.bashrc
```

## 推理

### C++版本推理

成功编译YoloV3工程后,在build目录下输入如下命令运行该示例:

```
./MIGraphX_Samples 0
```

程序运行结束会在build目录生成YoloV3检测结果图像。

<img src="./Resource/Images/Result.jpg" alt="Result" style="zoom:50%;" />

### python版本推理
Your Name's avatar
Your Name committed
77
78
79
80

YoloV3模型的推理示例程序是YoloV3_infer_migraphx.py,使用如下命令运行该推理示例:

```
Your Name's avatar
Your Name committed
81
82
83
84
85
86
87
# 进入python示例目录
cd ./Python

# 安装依赖
pip install -r requirements.txt

# 运行程序
Your Name's avatar
Your Name committed
88
89
90
91
92
93
94
95
python YoloV3_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0.4 \
	--confThreshold 置信度阈值,默认0.2 \
	--nmsThreshold nms阈值,默认0.4 \
```

Your Name's avatar
Your Name committed
96
程序运行结束会在当前目录生成YoloV3检测结果图像。
Your Name's avatar
Your Name committed
97

Your Name's avatar
Your Name committed
98
<img src="./Resource/Images/Result.jpg" alt="Result_2" style="zoom: 50%;" />
Your Name's avatar
Your Name committed
99
100
101
102
103
104
105
106

## 历史版本

​		https://developer.hpccube.com/codes/modelzoo/yolov3_migraphx

## 参考

​		https://github.com/ultralytics/yolov3