Deep_Learning & Neural_Network_Lab
实验环境
- Python 3.9
- Anaconda3(官网下载或清华镜像都行)
- PyTorch 1.10.1(安装代码如下)
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
数据集下载
- 实验使用MNIST数据集
- 下载地址: http://yann.lecun.com/exdb/mnist/
运行方式
- 启动PyTorch环境
conda activate pytorch
- 在该环境下运行程序(30-60 minutes per file)
python GAN.py # 运行GAN.py
python CGAN.py # 运行CGAN.py
python DCGAN.py # 运行DCGAN.py
python WGAN.py # 运行WGAN.py
python Improved_WGAN.py # 运行Improved_WGAN.py
实验结果
| Model | epoch | 1 | 5 | 10 | 50 | 100 | 150 | 200 |
|---|---|---|---|---|---|---|---|---|
| GAN | d_loss | 0.510469 | 0.342887 | 0.248609 | 0.170712 | 0.120419 | 0.169989 | 0.192855 |
| g_loss | 0.956933 | 1.874407 | 2.537149 | 3.263798 | 3.905378 | 3.570554 | 3.132934 | |
| CGAN | d_loss | 0.106481 | 0.135711 | 0.194618 | 0.17816 | 0.124881 | 0.106104 | 0.097463 |
| g_loss | 0.660399 | 0.585134 | 0.415263 | 0.474954 | 0.632873 | 0.689216 | 0.718078 | |
| DCGAN | d_loss | 0.692203 | 0.631018 | 0.588311 | 0.463967 | 0.334903 | 0.242517 | 0.103577 |
| g_loss | 0.694127 | 0.829934 | 0.936831 | 1.402277 | 2.230488 | 2.936166 | 2.757655 | |
| WGAN | d_loss | -0.291029 | -0.329138 | -0.468464 | -0.214993 | -0.123098 | -0.080621 | -0.060724 |
| g_loss | -12.6155 | -0.209976 | -0.662237 | -0.456756 | -0.427529 | -0.388925 | -0.197952 | |
| Improved_WGAN | d_loss | -0.246925 | -7.846449 | -6.713724 | -3.10867 | -2.067324 | -1.634501 | -1.393292 |
| g_loss | -14.636739 | 0.762477 | 1.745449 | -2.624076 | -2.965248 | -2.781996 | -2.585317 |
- 具体数据详见 data.xls 文件
