This is a project based on caffe and MSRA's Deep Image Analogy
I modified the project Deep Image Analogy from C version to python version(except photo transfer part). My work was included in ./deep_image_analogy/src/*
It seems that this work is not so meaningful to the real world, but I regardeded it as a personal project for me to have a taste of multiple technique skills, such as (py)caffe, deep learning, cuda, google cloud platform, django, git, etc.
Besides the technique skills, the soft skill -- problem solving was also trained through this project. I encountered a variety of problems and solved them one by one. For example, since this project needed a GPU to accelerate calculation while I did not have one and one NVIDIA GPU was quite expensive for me, but I did not give up and actively to seek a solution, and finally found Google Could Platform and rent a GPU(Tesla K80). These obstacles often appeared in my project, either big or small, I confronted them bravely.
In addition, I am a greener in github, if there is anything involved with your copyright, please tell me. I will delete it soon. Thanks a lot!
Hardware Enviroment(Google Cloud Platfrom):
• n1-standard-4 (4 vCPUs, 15 GB memory);
• 1 NVIDIA Tesla K80 (12 GB memory);
• 50 GB disk.
Software Enviroment(Google Cloud Platform):
• 64-bit Ubuntu 16.04 Operating System;
• NVIDIA Driver 396.36 + CUDA Toolkit 9.2.88 + cuDNN v7.1.4 for CUDA9.2 + pycuda;
• pycaffe;
• python 2.7;
• openCV 3.0.
Every two lines is a contrast combination. The first line is the result from original C version code, the second line is the result from my python code.
About Run Time: 70s(original program) VS 170s(my program)
From the result picture and running time, we can see that my code need improvement.
My codes acknowledge Deep Image Analogy and caffe