The repo contains refactored and compiled code implementing GANs using MATLAB GPU Coder. The code is provided in form of *.m scripts that implement GANs and compiled versions in form of *.MEX files. Compiled versions directly utilize CUDA kernels to speed up computation.

D. Coufal, F. Hakl, P. Vidnerová. The Czech Academy of Sciences, Institute of Computer Science

MATLAB GPU Coder, CUDA, generative adversial networks

MATLAB, MATLAB GPU Coder

• CUDA compiled code
• speeding up generation phase of GANs

- nck_dcgan.m, nck_lsgan - GANs implemented in MATLAB, parameters are set within the scripts
- *_generate.m - functions to generate rxc images from gan, r,c are parameters
- *_generate_mex.mexw64 - compiled versions for r=100, c_100; generate 10000 images

for uncompiled, CPU version generation of 10k images takes about 10 sec

tic;*_generate;toc

for compiled, CUDA native version, generation of 10k images takes about 1 sec

tic;*_generate_mex;toc

Consider to run the above commands at least twice to get stable times. The first run is usually not reliable.

This work was partially supported by the TAČR grant TN01000024 and institutional support of the Institute of Computer Science RVO 67985807