PPGN代码安装测试

即插即用神经网络

本笔记记录在ubuntu14.10+python2.7下的caffe环境配置和跑通全部PPGN测试样例

论文Plug & Play Generative Networks: Conditional Iterative Generation of Images in Latent Space

源码https://github.com/Evolving-AI-Lab/ppgn

1. Setup

1.1 Installing software

This code is built on top of Caffe. You'll need to install the following:

1.1.1 Install Caffe; follow the official installation instructions.

见笔记http://blog.csdn.net/lien0906/article/details/46816273

1.1.2 Build the Python bindings for Caffe

If you want to try example 5 (image captioning), you would need to use the Caffe provided here instead

这个意思是用绑定了python的caffe包代替一般的caffe。

1.1.3 You can optionally build Caffe with the GPU option to make it run faster (recommended)

将1.1.2中下载的caffe_lrcn-master文件夹中的Makefile.config.example原地赋值一份并重命名为Makefile.config，打开该文件，将CPU_ONLY := 1前面的#去掉（也就是反注释），选择只用cpu

1.1.4 Make sure the path to your caffe/python folder in settings.py is correct

settings.py中的caffe_root="/home/bigbug/caffe_lrcn-master"

1.1.5 Install ImageMagick command-line interface on your system (for post-processing the images)

ubuntu系统中默认安装了ImageMagick，可以通过ubuntu终端输入identify -version查看版本信息

1.2 Downloading models

1.2.1 You will need to download a few models to run the examples below. There are download.sh scripts provided for your convenience.

cd到对应的.sh文件回车即可。

generator.caffemodel

bvlc_reference_caffenet.caffemodel

placesCNN.tar.gz

lrcn_caffenet_iter_110000.caffemodel

• The generator network (Noiseless Joint PPGN-h) can be downloaded via: cd nets/generator/noiseless && ./download.sh
• The encoder network (here BVLC reference CaffeNet): cd nets/caffenet && ./download.sh
• For example 4, download AlexNet CNN trained on MIT Places dataset: cd nets/placesCNN && ./download.sh
• For example 5, download LRCN image captioning model: cd nets/lrcn && ./download.sh

1.2.2 Settings:

• Paths to the downloaded models are in settings.py. They are relative and should work if the download.sh scripts run correctly.

settings.py中的caffe_root="/home/bigbug/caffe_lrcn-master"，其他路径在默认下载之后保持不变。

2. Usage

Examples

We provide here 5 different examples as a starting point. Feel free to fork away to produce even cooler results!

1_class_conditional_sampling.sh: Sampling conditioning on the class "junco" (output unit #13 of the CaffeNet DNN trained on ImageNet dataset). This script produces a sampling chain for a single given class.

• Running ./1_class_conditional_sampling.sh 13

注意：这个.sh文件需要一个参数，例如13

另外，如果运行报错ImportError no module named caffe，说明python和caffe的接口没有设置好，方法为：

1）、更改配置文件

sudo vim ~/.bashrc

打开的文件末尾输入两行

#caffe

export PYTHONPATH=/home/bigbug/caffe_lrcn-master/python

保存

source ~/.bashrc

2）、安装其他接口模块

先安装

sudo apt-get install python-numpy python-scipy python-matplotlib python-sklearn python-skimage python-h5py python-protobuf python-leveldb python-networkx python-nose python-pandas python-gflags Cython ipython

再更新

sudo apt-get update

最后在/home/bigbug/caffe_lrcn-master路径下执行

make pycaffe -j8

3)、测试

在终端输入python进入python环境，输入命令import caffe，没有出现报错即可，再输入exit()返回ubuntu终端