这篇文章通过一个具体的例子来看一下使用torch可以做的有趣的事情。

事前准备

• https://liumiaocn.blog.csdn.net/article/details/109848433

liumiaocn:fast-neural-style liumiao$ th
 
  ______             __   |  Torch7 
 /_  __/__  ________/ /   |  Scientific computing for Lua. 
  / / / _ \/ __/ __/ _ \  |  Type ? for help 
 /_/  \___/_/  \__/_//_/  |  https://github.com/torch 
                          |  http://torch.ch 
	
th> 

步骤1: 下载图像风格迁移代码

执行 命令：git clone https://github.com/jcjohnson/fast-neural-style

liumiaocn:~ liumiao$ git clone https://github.com/jcjohnson/fast-neural-style
Cloning into 'fast-neural-style'...
remote: Enumerating objects: 3, done.
remote: Counting objects: 100% (3/3), done.
remote: Compressing objects: 100% (3/3), done.
remote: Total 256 (delta 0), reused 0 (delta 0), pack-reused 253
Receiving objects: 100% (256/256), 18.65 MiB | 7.14 MiB/s, done.
Resolving deltas: 100% (109/109), done.
liumiaocn:~ liumiao$ 

步骤2: 下载预先训练好的模型

liumiaocn:~ liumiao$ cd fast-neural-style/
liumiaocn:fast-neural-style liumiao$ ls
README.md		fast_neural_style.lua	print_options.lua	slow_neural_style.lua	webcam_demo.lua
doc			images			requirements.txt	test
fast_neural_style	models			scripts			train.lua
liumiaocn:fast-neural-style liumiao$ sh models/download_style_transfer_models.sh 
 ...省略
liumiaocn:fast-neural-style liumiao$ find models/ -name '*.t7'
models//instance_norm/mosaic.t7
models//instance_norm/candy.t7
models//instance_norm/the_scream.t7
models//instance_norm/feathers.t7
models//instance_norm/la_muse.t7
models//instance_norm/udnie.t7
models//eccv16/starry_night.t7
models//eccv16/la_muse.t7
models//eccv16/composition_vii.t7
models//eccv16/the_wave.t7
liumiaocn:fast-neural-style liumiao$ 

注意：此步骤中下载脚本可能会下载失败，可以通过其他方式比如浏览器下载之后移动到相应位置即可。

结果验证

示例1: 黑猫警长

• 原始图片
  
• 执行命令：

liumiaocn:fast-neural-style liumiao$ th fast_neural_style.lua \
>   -model models//eccv16/starry_night.t7 \
>   -input_image liumiaocn.jpg \
>   -output_image newstyle-liumiaocn.png
Writing output image to newstyle-liumiaocn.png	
liumiaocn:fast-neural-style liumiao$

• 风格变换之后
  
  注：图像太小，拉伸后很诡异

示例2: 芝加哥

• 原始图片
  
• 执行命令

liumiaocn:fast-neural-style liumiao$ time -p th fast_neural_style.lua \
>   -model models//eccv16/starry_night.t7 \
>   -input_image city.jpg \
>   -output_image newstyle-city.png
Writing output image to newstyle-city.png	
real 7.60
user 9.32
sys 4.79
liumiaocn:fast-neural-style liumiao$ 

• 风格变换之后