README.md 1.36 KB
Newer Older
sunxx1's avatar
sunxx1 committed
1
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
# InceptionV3

## 模型介绍
InceptionV3是一种卷积神经网络模型,由Google的研究员Christian Szegedy等人在2015年提出。该模型采用了一种名为“Inception”的模块化结构,允许网络在不同的尺度上进行卷积操作,从而提高了网络的精度和效率。

## 模型结构

InceptionV3的整体结构由多个Inception模块组成,每个Inception模块由多个并行的卷积层和池化层组成。这些并行的卷积层和池化层的输出再被级联在一起,形成了一个高维度的特征向量,这个向量再被输入到下一个Inception模块中进行处理。

## 数据集
在本测试中可以使用ImageNet数据集。

## InceptionV3训练

### 环境配置

提供[光源](https://www.sourcefind.cn/#/service-details)拉取的训练以及推理的docker镜像:

* 训练镜像:docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.10.0-centos7.6-dtk-22.10.1-py37-latest

## 训练

将训练数据解压到data目录下。

训练命令:

    ./inception_v3.sh ${isperf} ${output_json} ${ngpu}

sunxx1's avatar
sunxx1 committed
29
## 准确率数据
sunxx1's avatar
sunxx1 committed
30
31
32

测试数据使用的是ImageNet数据集,使用的加速卡是DCU Z100L。

sunxx1's avatar
sunxx1 committed
33
34
35
36
| 卡数 |           精度            |
| :--: | :-----------------------: |
|  8   | top1:0.73888;top5:0.90528 |

sunxx1's avatar
sunxx1 committed
37
### 源码仓库及问题反馈
sunxx1's avatar
sunxx1 committed
38
39

https://developer.hpccube.com/codes/modelzoo/inceptionv3_pytorch