contact: mzha (at) ust(dot) hk

Surrealism_DCGAN

Reimplementation of art-DCGAN by Robbie Barrat and Jordan Bird in art-dcgan-keras. Pics of style surrealism is scraped from WikiArt for training.

utils

utils is a folder for obtaining the training dataset from scratch. it is composed of:

• WikiArtScraper.py
• UrlDownload.py
• ImageProcessor.py

The dataset is scrped from WikiArt by WikiArtScraper.py and UrlDownload.py and then processed by ImageProcessor.py. The codes relies heavily on previous work down by CHEN Huangrong

data

data contains the pics scraped from WikiArt for training. Traing data for each task is named after the corresponding genre, for example, surrealism. Applying ImageProcessor.py to surrealism gives us a processed imageset surrealismCroped. Images in surrealismCroped is of the same size 128*128 of RGB coding.

output

output stores outputs in two subfolder

• samples
• weights

Training Data Examples

Outputs of Day 2

Future work

• Fire the art-dcgan-keras.py to command line with genre, epoch ect. as args
• Keep training model for finer outcome in surrealism (daily updating)