Skip to content

Visual Attribute Transfer through Deep Image Analogy in PyTorch!

License

Notifications You must be signed in to change notification settings

harveyslash/Deep-Image-Analogy-PyTorch

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Deep-Image-Analogy

Unofficial,PyTorch version of Deep Image Analogy.https://arxiv.org/abs/1705.01088. This project focuses on documentation of the project , and simplifying the structure. A blog post on it is coming soon.

Some Outputs (The two images in the middle are generated by the Algorithm)

Some other interesting (one way) outputs

This Project uses python3.6 and Cuda

To Install Dependencies:

be in the root directory and run pip install -r requirements.txt

To See Step by Step working of Project :

Run the Deep Image Analogy.ipynb file in the notebooks folder (using jupyter)

To run project:

cd into src , and run python Deep-Img-Analogy.py INPUT_IMG_A INPUT_IMG_BB OUTPUT_IMG

Note

This project uses Adam as optimizer instead of LBFGS. LBFGS was giving really poor results.

Project Organization

├── data
│   ├── outputs <-- folder to store outputs
│   └── raw <-- folder to store inputs
├── LICENSE.md
├── notebooks
│   ├── Deep Image Analogy.ipynb Full Pipeline in a step by step manner
│   ├── PatchMatch-Demo.ipynb Raw Patchmatch demo
│   └── WLS.ipynb Weighted Least Squares Implementation Demo (currently not being used by this project)
├── README.md 
├── requirements.txt <-- Project requirements. 
└── src
    ├── Deep-Img-Analogy.py <-- End to end executable with command line interface.
    ├── models
    │   └── VGG19.py <-- modified VGG19 with support for deconvolution, and other things. 
    ├── PatchMatch
    │   └── PatchMatchOrig.py <-- CPU version of PatchMatch. GPU version may come in the future.
    ├── Utils.py <-- Helper Utilities
    └── WLS.py <-- Weighted Least Squares.

Project based on the cookiecutter data science project template. #cookiecutterdatascience