README.md 2.12 KB
Newer Older
dcuai's avatar
dcuai committed
1
# MobileNetV3
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

## 论文

Searching for MobileNetV3

https://arxiv.org/pdf/1905.02244.pdf

## 模型结构

MobileNetv3模型采用轻量级的深度可分离卷积(depthwise separable convolution)结构,以减少模型参数量和计算复杂度。

![Backbone](Backbone.png)

## 算法原理

MobileNetv3模型采用混合使用轻量级深度可分离卷积和逆残差结构(Inverted Residuals)的算法原理,以实现高效计算和良好的模型性能

![Algorithm_principle](Algorithm_principle.png)

## 环境配置
### Docker(方法一)
此处提供[光源](https://www.sourcefind.cn/#/service-list)拉取docker镜像的地址与使用步骤

```
docker pull image.sourcefind.cn:5000/dcu/admin/base/custom:tvm_fp32_dtk22.10.1_py38_centos-7.6-latest
docker run -it -v /path/your_code_data/:/path/your_code_data/ --shm-size=32G --privileged=true --device=/dev/kfd --device=/dev/dri/ --group-add video --name docker_name imageID bash
cd /root/tvm-0.11-dev0/apps/
chenzk's avatar
chenzk committed
29
git clone  http://developer.sourcefind.cn/codes/modelzoo/mobilenet_v3_tvm.git
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
30
31
32
```

## 数据集
chenzk's avatar
chenzk committed
33
在本测试中可以使用ImageNet-2012数据集。
dengjb's avatar
update  
dengjb committed
34
35
36
- 官方下载通道
  - [https://image-net.org/](https://image-net.org/)
  数据结构如下:
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
```
data
    ├──imagenet
        ├── meta
            ├──val.txt
            ├──train.txt
            ...
        ├── train
        ├── val
```


## 推理

Python Deploy测试参考:
```
    cd  mobilenet_v3_tvm
    ulimit -s unlimited
    export MIOPEN_DEBUG_CONV_IMPLICIT_GEMM=0
    export HIP_VISIBLE_DEVICES=2(此处修改为对应加速卡ID号)
    python  val_onnx.py --test-only  --data-path /parastor/DL_DATA/ImageNet-pytorch --model mobilenet_v3_large --b 1 --pretrained
```
dcuai's avatar
dcuai committed
59
## result
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
60
61
![result](result.png)

zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
62
## 精度
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
63
64

测试数据:ImageNet-pytorch,使用的加速卡:Z100L。
65
66
67
| Acc@1 | Acc@5 |
| :------: | :------: |
| 74.054 | 91.340 |
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
68
69
70
71
72
## 应用场景
### 算法类别
图像分类

### 热点行业
dcuai's avatar
dcuai committed
73
74
制造,能源,交通,网安
## 源码仓库及问题反馈
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
75

chenzk's avatar
chenzk committed
76
https://developer.sourcefind.cn/codes/modelzoo/mobilenet_v3_tvm
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
77

dcuai's avatar
dcuai committed
78
## 参考资料
zhangqha@sugon.com's avatar
zhangqha@sugon.com committed
79
80

https://github.com/open-mmlab/mmpretrain
zhangqha's avatar
zhangqha committed
81