Skip to content

Using deep learning and generative adversarial networks to create novel human eyes

Notifications You must be signed in to change notification settings

williamjchen/eyeGAN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

EyeGAN

Generating human eyes with a DCGAN

Requirements

  • Tensorflow
  • Numpy
  • Python 3

Getting Started

eye dataset is the eyes (copy) folder. Around 1000 human and a few non-human eyes.

Run main.py to begin training.

Data

Using the images from this simliar project and bulk bing image downloader, I was able to amass around 1000 images of eyes. Using tensorflow to reflect each image horizontally, the dataset was augmented to ~200 images.

Results

*Note the ordering of the images are incorrect for the black and white images. I made an error naming the files. The epoch is incorrect.

Final_bw Final_c Gif Gif Gif

Reflection

The results of the experiment were arguablly good. The generated images are unquestionably eyes, and have resolution mimicing real eyes. Unfortunately, there are still many artifacts in the images, which are especially notable in the coloured ones. An increase of training data would likely increase image quality, but that proved difficult during the process. Many eyes that appear on a internet search are drawings or repeated. Future work should focus on increasing data size through other means like extracting eyes from larger images of people through object detection models like YOLO.

About

Using deep learning and generative adversarial networks to create novel human eyes

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages