HyperGAN
HyperGAN is a Generative Adversarial Network(GAN) with API and a command line tool.[1]HyperGAN's first stable release was on on Dec 25, 2016 and was version 0.1.[2]It was developed by 255BITS, namely, Mikkel Garcia, and Martyn Garcia. It is still releasing versions as of 06 February 2018.[3]The neural network is made for scalability and ease-of-use.[4]HyperGAN is also used to create computer generated imagery (or CGI).[5]
HyperGAN is written in Python, an easy-to-use programming language.
Imagery[edit]
HyperGAN uses training images to create CGI. To start the training, the user must first install HyperGAN. To begin running the training, the user must first compile a stockpile of images in a folder. After this, the user must input this command into a Python interperator, after importing HyperGAN to begin the training.
# Train a 32x32 gan with batch size 32 on a folder of folders of pngs, resizing images as necessary
hypergan train folder/ -s 32x32x3 -f png -c mymodel --resize
The # tells the code to look over this, and not run the command. The # (or comment) tells what HyperGAN will do when the code is ran. To save images a similar command is used. By default HyperGAN will not save images to the disk.
--save_samples.
This tells HyperGAN to save images to the disk automatically.
HyperGAN can also save the training data as a video or .mp4 file using the following command.[6]
ffmpeg -i samples/%06d.png -vcodec libx264 -crf 22 -threads 0 gan.mp4
In Popular Culture[edit]
The YouTuber by the name of Draft:carykh, a computer savvy programmer, has used HyperGAN multiple times to compose music, create images of human faces, create scenes from Cary's very own Battle for Dream Island, and try to replicate his voice. HyperGAN is not made to compose music or replicate human voices, as it only outputs images; however, Cary used another piece of software that he wrote in processing IDE to be able to listen to the music files along with ARSS (The Analysis & Resythesis Sound Spectograph) to listen t o the AI trying to replicate his voice. carykh has also pit three other neural networks to see which would compose music the best. HyperGAN was included, along with PixelCNN and CBLSTM.[7][8][9][10][11]
References[edit]
- ↑ 255BITS. "HyperGAN Official Site".
- ↑ 255BITS. "HyperGAN Releases".
- ↑ 255BITS. "255BITS Site and Projects".
- ↑ 255BITS. "255BITS/HyperGAN".
- ↑ Jolo Balbin (13 August 2017). "Using Generative Adversarial Networks (GAN) to Produce Different Faces of Crocodiles".
- ↑ 255BITS. "HyperGAN Training".
- ↑ carykh. "Teaching my computer to give me friends (I... I mean images!) (convolutional neural networks)".
- ↑ carykh. "AI makes new BFDI screenshots! (2 min. of ML & 22 min. of Cary whining about problems)".
- ↑ carykh. "3 neural nets battle to produce the best jazz music".
- ↑ carykh. "AI evoloves to compose three hours of jazz!".
- ↑ carykh. "Computer tries to replicate my voice!".
External links[edit]
This article "HyperGAN" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:HyperGAN. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.
