This is the official implementation for our paper presented at ICIP 2022:
A Patch-Based Algorithm for Diverse and High Fidelity Single Image Generation
N. Cherel, A. Almansa, Y. Gousseau, A. Newson
Link to [Preprint] [Paper]
We present a pure patch-based solution to single image generation that does not require learning. As a result new samples are possible using this code in a few seconds.
This algorithm contains the code for our PSin algorithm only.
The reference code for the optimal transport initialization is found at optimization https://github.com/ahoudard/wgenpatex . Our fork with minor modifications will be released soon.
| Reference | Generated |
|---|---|
 |
 |
The requirements are:
- opencv
- numpy
- scipy
- cffi
- numba
You can accelerate processing by compiling the source file patch_measure.cpp with the following command (tested on Linux only):
g++ -fPIC -shared patch_measure.cpp -O3 -o libpatch_measure.so
And then activate it in config.py with USE_CPP=True.
The code is then run using :
python synthesis.py
The default file used as reference is available is balloons.png.
