SinGAN_easy_tensorflow

Native TensorFlow API implementation of SinGAN paper (https://arxiv.org/pdf/1905.01164.pdf).

The goal of this implementation is instructional. I hope it is useful for people who do not have deep background in Pytorch or Keras. This implementation defines model graph using native TensorFlow (version lte 2.0), thus this may be easier to understand the method for beginners.

Contents

1. Results
2. Modification from original paper
3. Installation
4. Usage

Results

Image	Random Sample 1	Random Sample 2	Random Sample 3

Modification from original paper

In the paper, to ensure the original image exists in the solution space of generator, noise map is kept as (z^*_N, 0, 0, 0, ...) from coarsest scale to finest scale. Here z^*_N is kept fixed while training. Further random samples are generated from noise map (z_N, z_N-1, z_N-2, ...) etc.

We instead trained the network by generating random samples using the following map (z^*_N + z_N, z_N-1, z_N-2, ...). This way, the noise at the coarsest scale is centered around z^*_N. Thus, the random sample at coarsest scale can be made to preserve global pattern in images by tuning the magnitude of z_N. The resultant images at finest scale preserve global structure much better.

Figure below show comparisons for modified strategy (top) and original strategy (bottom)

Random Sample 1	Random Sample 2	Random Sample 3	Random Sample 4

Installation

Installation steps tested for Windows 10 64-bit and Python 3.5

1. Download repository

Open Git Bash terminal, navigate to desired location and clone repository using git clone https://github.com/shiveshc/SinGAN_easy_tensorflow.git.

Or click on Code button on top right corner and Download ZIP.

2. Set up venv and instal libraries

Open command line terminal as administrator and navigate to cloned repository path using cd .\SinGAN_easy_tensorflow.

Next run following commands -

Windows 10	MacOS
1. python -m venv env	1. python -m venv env
2. env\Scripts\activate.bat	2. source env/bin/activate
3. python -m pip install --upgrade "pip < 21.0"	3. python -m pip install --upgrade "pip < 21.0"
4. pip install -r requirements.txt	4. pip install -r requirements.txt

Installation should take 10-15 minutes.

Common installation errors -

1. pip version is not upgraded
   solution : upgrade pip using python -m pip install --upgrade pip
2. pip version is not compatible with python version
   solution : install suitable pip version e.g. pip version < 21.0 is compatible with Python 3.5.

Usage

To train on balloons.png in data folder run - \

1. env\Scripts\activate.bat\
2. python train.py /data/balloons.png output_path -scales 6 -iters 75 1