CycleGAN（三）代码概览

矫情吗；* 2022-02-27 02:08 780阅读 0赞

**目的**：大致看懂cycleGAN代码结构

**参考**：[https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/docs/overview.md][https_github.com_junyanz_pytorch-CycleGAN-and-pix2pix_blob_master_docs_overview.md]

**目录**

一、概览

train.py

test.py

data文件夹

models文件夹

options文件夹

util文件夹

二、train.py

三、test.py

四、data文件夹

4.1 \_\_init\_\_.py

4.2 base\_dataset.py

4.3 image\_folder.py

4.4 template\_dataset.py

4.5 aligned\_dataset.py

4.6 unaligned\_dataset.py

4.7 single、clolorization dataset

五、models

六、options and util

--------------------

# 一、概览 #

### train.py ###

用于模型训练

`--model`: e.g., `pix2pix`, `cyclegan`, `colorization`) and

different datasets (with option `--dataset_mode`: e.g., `aligned`, `unaligned`, `single`, `colorization`

### test.py ###

用于模型测试

### data文件夹 ###

包含关于所有数据加载数据处理的程序。

### models文件夹 ###

模型相关的objective functions, optimizations, and network architectures.

### options文件夹 ###

训练，测试以及相关模型的选项

### util文件夹 ###

相关帮助函数的杂项汇总

# 二、train.py #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70][]

train获取数据（data文件夹），创建模型（model），循环epoch然后在epoch内更新参数，存储网络

# 三、test.py #

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 1][]

读取数据，创建模型，然后将数据送入模型进行test

# 四、data文件夹 #

包含着模型加载与处理的程序。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 2][]

## 4.1 \_\_init\_\_.py ##

用于给train和test过程生成数据集

from data import create_dataset
    dataset = create_dataset(opt) to create a dataset given the option opt.

## 4.2 base\_dataset.py ##

[https://docs.python.org/3/library/abc.html][https_docs.python.org_3_library_abc.html]

*  [base\_dataset.py][base_dataset.py] implements an abstract base class ([ABC][https_docs.python.org_3_library_abc.html]) for datasets. It also includes common transformation functions (e.g., `get_transform`, `__scale_width`), which can be later used in subclasses.

用于运用abstract base class abc。

## 4.3 image\_folder.py ##

pytorch默认只从文件夹中读文件，作者可以从文件夹和子文件夹中读文件。

## 4.4 template\_dataset.py ##

创建一个数据集的模板，以及详细的描述

## 4.5 aligned\_dataset.py ##

用于加载样本对（主要用于pix2pix，对于我们的cycleGAN并无太大作用）

## 4.6 unaligned\_dataset.py ##

用于unpaired 数据集，用于cycleGAN，训练时trainA 和trainB 中应该放入domainA和domainB中的东西，test时也是这样。

##  4.7 single、clolorization dataset ##

*  [image\_folder.py][image_folder.py] implements an image folder class. We modify the official PyTorch image folder [code][] so that this class can load images from both the current directory and its subdirectories.
 *  [template\_dataset.py][template_dataset.py] provides a dataset template with detailed documentation. Check out this file if you plan to implement your own dataset.
 *  [aligned\_dataset.py][aligned_dataset.py] includes a dataset class that can load image pairs. It assumes a single image directory `/path/to/data/train`, which contains image pairs in the form of \{A,B\}. See [here][] on how to prepare aligned datasets. During test time, you need to prepare a directory `/path/to/data/test` as test data.
 *  [unaligned\_dataset.py][unaligned_dataset.py] includes a dataset class that can load unaligned/unpaired datasets. It assumes that two directories to host training images from domain A `/path/to/data/trainA` and from domain B `/path/to/data/trainB` respectively. Then you can train the model with the dataset flag `--dataroot /path/to/data`. Similarly, you need to prepare two directories `/path/to/data/testA` and `/path/to/data/testB` during test time.
 *  [single\_dataset.py][single_dataset.py] includes a dataset class that can load a set of single images specified by the path `--dataroot /path/to/data`. It can be used for generating CycleGAN results only for one side with the model option `-model test`.
 *  [colorization\_dataset.py][colorization_dataset.py] implements a dataset class that can load a set of nature images in RGB, and convert RGB format into (L, ab) pairs in [Lab][] color space. It is required by pix2pix-based colorization model (`--model colorization`).

# 五、models #

modules related to objective functions, optimizations, and network architectures.

更改models的顺序：

To add a custom model class called 'dummy', you need to add a file called 'dummy\_model.py' and define a subclass DummyModel inherited from BaseModel.  
You need to implement the following five functions:  
    -- <\_\_init\_\_>:                      initialize the class; first call BaseModel.\_\_init\_\_(self, opt).  
    -- <set\_input>:                     unpack data from dataset and apply preprocessing.  
    -- <forward>:                       produce intermediate results.  
    -- <optimize\_parameters>:           calculate loss, gradients, and update network weights.  
    -- <modify\_commandline\_options>:    (optionally) add model-specific options and set default options.

In the function <\_\_init\_\_>, you need to define four lists:  
    -- self.loss\_names (str list):          specify the training losses that you want to plot and save.  
    -- self.model\_names (str list):         define networks used in our training.  
    -- self.visual\_names (str list):        specify the images that you want to display and save.  
    -- self.optimizers (optimizer list):    define and initialize optimizers. You can define one optimizer for each network. If two networks are updated at the same time, you can use itertools.chain to group them. See cycle\_gan\_model.py for an usage.

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 3][]

# 六、options and util #

![20190326204552305.png][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 4][]

[https_github.com_junyanz_pytorch-CycleGAN-and-pix2pix_blob_master_docs_overview.md]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/docs/overview.md
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70]: /images/20220227/6d176da84a4249ecad6091f3b3e9f5ad.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 1]: /images/20220227/07b3673b1d4843cb83feb4a568511ce7.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 2]: /images/20220227/adc8e6fc928647d39c7dcc028eb25b0d.png
[https_docs.python.org_3_library_abc.html]: https://docs.python.org/3/library/abc.html
[base_dataset.py]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/data/base_dataset.py
[image_folder.py]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/data/image_folder.py
[code]: https://github.com/pytorch/vision/blob/master/torchvision/datasets/folder.py
[template_dataset.py]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/data/template_dataset.py
[aligned_dataset.py]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/data/aligned_dataset.py
[here]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/docs/tips.md#prepare-your-own-datasets-for-pix2pix
[unaligned_dataset.py]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/data/unaligned_dataset.py
[single_dataset.py]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/data/single_dataset.py
[colorization_dataset.py]: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix/blob/master/data/colorization_dataset.py
[Lab]: https://en.wikipedia.org/wiki/CIELAB_color_space
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 3]: /images/20220227/936868d17213468596851b91686b1b6d.png
[20190326204552305.png]: /images/20220227/0fea2a9d26264424a9469580bfc99038.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl8zNjQ3NDgwOQ_size_16_color_FFFFFF_t_70 4]: /images/20220227/1c36331b8f8f4b98a823086992753772.png