本项目基于PaddlePaddle复现Wide Resnet，它是ResNet变体，改进了shortcut，采用更宽卷积并加dropout层。在CIFAR10测试集精度达96.6%，提供单卡和多卡训练方式，代码含模型、训练、评估等文件，依赖PaddlePaddle≥2.0.0，支持GPU和CPU运行。

一、简介

本项目基于paddlepaddle框架复现Wide Resnet，他是resnet的一种变体，主要区别在于对resnet的shortcut进行了改进，使用更“宽”的卷积以及加上了dropout层。

论文:

• [1] Zagoruyko S , Komodakis N . Wide Residual Networks[J]. 2016.
• 链接：Wide Residual Networks

参考项目：

• https://github.com/xternalz/WideResNet-pytorch
• https://github.com/meliketoy/wide-resnet.pytorch/blob/master/networks/wide_resnet.py

二、复现精度

该列指标在cifar10的测试集测试

train from scratch细节：

epochoptbatch_sizedatasetmemorycardprecision1400SGD128CIFAR1016G10.9660

模型下载?模型地址：aistudio

三、数据集

CIFAR10数据集。

• 数据集大小：
  • 训练集：50000张
  • 测试集：10000张
  • 尺寸：32 * 32
• 数据格式：分类数据集

四、环境依赖

• 硬件：GPU、CPU

• 框架：

  • PaddlePaddle >= 2.0.0

五、快速开始

step1: clone

# clone this repogit clone https://github.com/PaddlePaddle/Contrib.gitcd wide_resnetexport PYTHONPATH=./登录后复制

安装依赖

python3 -m pip install -r requirements.txt登录后复制

step2: 训练

python3 train.py登录后复制登录后复制

如果你想分布式训练并使用多卡：

python3 -m paddle.distributed.launch --log_dir=./debug/ --gpus '0,1,2,3' train.py登录后复制登录后复制

此时的输出为：

Epoch 0: PiecewiseDecay set learning rate to 0.05.iter:0 loss:2.4832iter:10 loss:2.3544iter:20 loss:2.3087iter:30 loss:2.2509iter:40 loss:2.2450登录后复制

step3: 测试

python3 eval.py登录后复制登录后复制

此时的输出为：

acc:9660 total:10000 ratio:0.966登录后复制登录后复制

六、代码结构与详细说明

6.1 代码结构

│ wide_resnet.py # 模型文件│ eval.py # 评估│ README.md # 英文readme│ README_cn.md # 中文readme│ requirement.txt # 依赖│ train.py # 训练登录后复制

6.2 参数说明

无

6.3 训练流程

单机训练

python3 train.py登录后复制登录后复制

多机训练

python3 -m paddle.distributed.launch --log_dir=./debug/ --gpus '0,1,2,3' train.py登录后复制登录后复制

此时，程序会将每个进程的输出log导入到./debug路径下：

.├── debug│ ├── workerlog.0│ ├── workerlog.1│ ├── workerlog.2│ └── workerlog.3├── README.md└── train.py登录后复制

训练输出

执行训练开始后，将得到类似如下的输出。每一轮batch训练将会打印当前epoch、step以及loss值。

Epoch 0: PiecewiseDecay set learning rate to 0.05.iter:0 loss:2.4832iter:10 loss:2.3544iter:20 loss:2.3087iter:30 loss:2.2509iter:40 loss:2.2450登录后复制

6.4 评估流程

python3 eval.py登录后复制登录后复制

此时的输出为：

acc:9660 total:10000 ratio:0.966登录后复制登录后复制

七、模型信息

关于模型的其他信息，可以参考下表：

信息说明发布者徐铭远时间2021.08框架版本>=Paddle 2.0.2应用场景图像分类支持硬件GPU、CPU下载链接预训练模型In [?]

# 以下为在aistudio上直接运行登录后复制In [4]

# 训练!python3 train.py登录后复制

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations def convert_to_list(value, n, name, dtype=np.int):W0808 16:41:54.148313 32483 device_context.cc:362] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 10.1, Runtime API Version: 10.1W0808 16:41:54.152312 32483 device_context.cc:372] device: 0, cuDNN Version: 7.6.Epoch 0: PiecewiseDecay set learning rate to 0.05.iter:0 loss:2.4279iter:10 loss:2.3434^C登录后复制In [?]

# 评估!python3 eval.py登录后复制

/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations def convert_to_list(value, n, name, dtype=np.int):W0808 16:37:21.490298 32096 device_context.cc:362] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 10.1, Runtime API Version: 10.1W0808 16:37:21.494925 32096 device_context.cc:372] device: 0, cuDNN Version: 7.6.acc:9660 total:10000 ratio:0.966登录后复制