README.md 3.02 KB
Newer Older
yangql's avatar
yangql committed
1
# ResNet50
yangql's avatar
yangql committed
2
3
## 论文
`Deep Residual Learning for Image Recognition`
yangql's avatar
yangql committed
4

yangql's avatar
yangql committed
5
- https://arxiv.org/abs/1512.03385
yangql's avatar
yangql committed
6
7
8
9

## 模型结构
ResNet50模型包含了49个卷积层、一个全连接层。

yangql's avatar
yangql committed
10
![ResNet50模型结构](./Doc/images/1.png)
yangql's avatar
yangql committed
11

yangql's avatar
yangql committed
12
13
## 算法原理
ResNet50使用了多个具有残差连接的残差块来解决梯度消失或梯度爆炸问题,并使得网络可以向更深层发展。
yangql's avatar
yangql committed
14

yangql's avatar
yangql committed
15
![算法原理](./Doc/images/2.png)
yangql's avatar
yangql committed
16

yangql's avatar
yangql committed
17
## 环境配置
yangql's avatar
yangql committed
18
### Docker(方法一)
yangql's avatar
yangql committed
19
拉取镜像:
yangql's avatar
yangql committed
20
21
22
23
```python
docker pull image.sourcefind.cn:5000/dcu/admin/base/custom:ort1.14.0_migraphx3.0.0-dtk22.10.1
```

yangql's avatar
yangql committed
24
创建并启动容器:
yangql's avatar
yangql committed
25
```
yangql's avatar
yangql committed
26
docker run --shm-size 16g --network=host --name=resnet50_ort --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v $PWD/resnet50_onnxruntime:/home/resnet50_onnxruntime -it <Your Image ID> /bin/bash
yangql's avatar
yangql committed
27
28
29

# 激活dtk
source /opt/dtk/env.sh
yangql's avatar
yangql committed
30
```
yangql's avatar
yangql committed
31
32
33
34
35
36
37
### Dockerfile(方法二)
```
cd ./docker
docker build --no-cache -t resnet50_onnxruntime:2.0 .

docker run --shm-size 16g --network=host --name=resnet50_ort --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v $PWD/resnet50_onnxruntime:/home/resnet50_onnxruntime -it <Your Image ID> /bin/bash
```
yangql's avatar
yangql committed
38

yangql's avatar
yangql committed
39
40
## 数据集
Imagenet-1k
yangql's avatar
yangql committed
41

yangql's avatar
yangql committed
42
43
44
45
## 推理
### Python版本推理
采用ONNXRuntime框架使用DCU进行推理,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。
#### 运行示例
yangql's avatar
yangql committed
46
```python
yangql's avatar
yangql committed
47
48
# 进入resnet50 onnxruntime工程根目录
cd <path_to_resnet50_onnxruntime> 
yangql's avatar
yangql committed
49

yangql's avatar
yangql committed
50
51
52
# 下载模型
sh ./Resource/Models/download.sh

yangql's avatar
yangql committed
53
54
55
56
57
# 进入示例程序目录
cd Python/

# 安装依赖
pip install -r requirements.txt
yangql's avatar
yangql committed
58

yangql's avatar
yangql committed
59
# 运行示例
yangql's avatar
yangql committed
60
61
62
python Classifier.py
```

yangql's avatar
yangql committed
63
### C++版本推理
yangql's avatar
yangql committed
64
采用ONNXRuntime框架使用DCU进行推理,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。
yangql's avatar
yangql committed
65
#### 构建工程
yangql's avatar
yangql committed
66
```
yangql's avatar
yangql committed
67
68
69
70
cd <path_to_resnet50_onnxruntime>

# 下载模型
sh ./Resource/Models/download.sh
yangql's avatar
yangql committed
71

yangql's avatar
yangql committed
72
# 安装Opencv依赖
yangql's avatar
yangql committed
73
sh ./3rdParty/InstallOpenCVDependences.sh
yangql's avatar
yangql committed
74

yangql's avatar
yangql committed
75
76
rbuild build -d depend
```
yangql's avatar
yangql committed
77
#### 设置环境变量
yangql's avatar
yangql committed
78
79
将依赖库依赖加入环境变量LD_LIBRARY_PATH,在~/.bashrc中添加如下语句:
```
yangql's avatar
yangql committed
80
export LD_LIBRARY_PATH=<path_to_resnet50_onnxruntime>/depend/lib64/:$LD_LIBRARY_PATH
yangql's avatar
yangql committed
81
82
83
84
85
```
然后执行:
```
source ~/.bashrc
```
yangql's avatar
yangql committed
86

yangql's avatar
yangql committed
87
88
#### 运行示例
```
yangql's avatar
yangql committed
89
# 进入resnet50_onnxruntime工程根目录
yangql's avatar
yangql committed
90
cd <path_to_resnet50_onnxruntime> 
yangql's avatar
yangql committed
91

yangql's avatar
yangql committed
92
93
94
#下载模型
sh ./Resource/Models/download.sh

yangql's avatar
yangql committed
95
96
97
98
99
100
# 进入build目录
cd build/

# 执行示例程序
./ResNet50
```
yangql's avatar
yangql committed
101
## result
yangql's avatar
yangql committed
102
103
104
105
106
107
108
109
### python版本
```
class=n01843065 jacamar ; probability=11.903831
```
### C++版本
```
class=n01843065 jacamar ; probability=11.903831
```
yangql's avatar
yangql committed
110
111
### 精度

yangql's avatar
yangql committed
112

yangql's avatar
yangql committed
113
114
115
## 应用场景
### 算法类别
`图像分类`
yangql's avatar
yangql committed
116

yangql's avatar
yangql committed
117
118
119
120
121
### 热点应用行业
`制造,政府,医疗,科研`

## 源码仓库及问题反馈
https://developer.hpccube.com/codes/modelzoo/resnet50_onnxruntime
yangql's avatar
yangql committed
122

yangql's avatar
yangql committed
123
## 参考资料
yangql's avatar
yangql committed
124
125
https://github.com/onnx/models/tree/main/vision/classification/resnet