Commit 2c975067 authored by chenzk's avatar chenzk
Browse files

v1.2.4

parent 5082bb1e
Pipeline #600 canceled with stage
......@@ -3,14 +3,19 @@
## 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。
<div align=center>
<img src="./doc/vit.png"/>
</div>
![img](./doc/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:
<div align=center>
<img src="./doc/attention.png"/>
</div>
![img](./doc/attention.png)
## 环境配置
```
mv megatron-deepspeed-vit_pytorch megatron-deepspeed-vit # 去框架名后缀
......@@ -99,12 +104,14 @@ sh examples/dspvit_1dcu.sh
sh examples/dspvit_1node.sh
```
## result
![img](./doc/classify.png)
<div align=center>
<img src="./doc/classify.png"/>
</div>
## 应用场景
### 算法类别
`图像分类`
### 热点应用行业
`制造,环,医疗,气象`
`制造,环,医疗,气象`
## 源码仓库及问题反馈
- https://developer.hpccube.com/codes/modelzoo/megatron-deepspeed-vit_pytorch
## 参考资料
......
......@@ -5,6 +5,6 @@ modelName=megatron-deepspeed-vit_pytorch
# 模型描述
modelDescription=基于transformer的图像分类算法
# 应用场景
appScenario=推理,训练,图像分类,制造,环,医疗,气象
appScenario=推理,训练,图像分类,制造,环,医疗,气象
# 框架类型
frameType=PyTorch
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment