README.md

# ViT 

## 论文
`An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale`
- https://arxiv.org/abs/2010.11929
## 模型结构
Vision Transformer先将图像用卷积进行分块以降低计算量，再对每一块进行展平处理变成序列，然后将序列添加位置编码和cls token，再输入多层Transformer结构提取特征，最后将cls tooken取出来通过一个MLP（多层感知机）用于分类。

![img](./docs/vit.png)
## 算法原理
图像领域借鉴《Transformer is all you need！》算法论文中的Encoder结构提取特征，Transformer的核心思想是利用注意力模块attention提取特征：

![img](./docs/attention.png)
## 环境配置
### Docker（方法一）
```
docker pull image.sourcefind.cn:5000/dcu/admin/base/custom:decode-ffmpeg-dtk23.04
# <your IMAGE ID>用以上拉取的docker的镜像ID替换
docker run --shm-size 10g --network=host --name=vit_migraphx --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v path_to_vit_migraphx:/home/vit_migraphx -it <your IMAGE ID> bash
```
### Dockerfile（方法二）
```
cd vit_migraphx/docker
docker build --no-cache -t vit_migraphx:test .
docker run --rm --shm-size 10g --network=host --name=vit_migraphx --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v path_to_vit_migraphx:/home/vit_migraphx -it vit_migraphx:test bash
```
## 数据集

下载推理数据
```
wget https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz
tar -zxvf flower_photos.tgz
数据结构如下：
flower_photos
├── daisy
│   ├── 100080576_f52e8ee070_n.jpg
│   ├── 10140303196_b88d3d6cec.jpg
│   └── 99306615_739eb94b9e_m.jpg
├── dandelion
│   ├── 10043234166_e6dd915111_n.jpg
│   ├── 10200780773_c6051a7d71_n.jpg
│   └── 9965757055_ff01b5ee6f_n.jpg
├── LICENSE.txt
├── roses
│   ├── 10090824183_d02c613f10_m.jpg
│   ├── 102501987_3cdb8e5394_n.jpg
│   └── 99383371_37a5ac12a3_n.jpg
├── sunflowers
│   ├── 1008566138_6927679c8a.jpg
│   ├── 1022552002_2b93faf9e7_n.jpg
│   └── 9904127656_f76a5a4811_m.jpg
└── tulips
    ├── 100930342_92e8746431_n.jpg
    ├── 10094729603_eeca3f2cb6.jpg
    └── 9976515506_d496c5e72c.jpg
```

## 推理

### 编译
```
git clone https://developer.hpccube.com/codes/modelzoo/vit_migraphx.git
cd vit_migraphx
make
```
### 执行推理
```
./ViT_MIGraphX
```
根据提示选择要运行的示例程序，比如执行
```
./ViT_MIGraphX --models=Models/model.onnx --input=flower_photos/daisy/
```
运行ViT模型，对daisy图片进行分类


## 精度

测试数据使用的是[flower_photos](https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz)，使用的加速卡是DCU Z100

| Engine | Model Path| Data | Accuracy(%) |
| :------: | :------: | :------: | :------: |
| MIGraphX | models/model.onnx | daisy | 98.4 |
| MIGraphX | models/model.onnx | dandelion | 98.1 |
| MIGraphX | models/model.onnx | roses | 91.3 |
| MIGraphX | models/model.onnx | sunflowers | 97.4 |
| MIGraphX | models/model.onnx | tulips | 94.1 |


## 应用场景
### 算法类别
`图像分类`
### 热点应用行业
`制造,环境,医疗,气象`
## 源码仓库及问题反馈
- https://developer.hpccube.com/codes/modelzoo/vit_migraphx.git
## 参考资料
- https://github.com/WZMIAOMIAO/deep-learning-for-image-processing