CGAN-Emotion-Art

Project overview

• From Wiki-Art Emotion, We built a VGG16 Classifier (98% train acc - 50 val acc) with 3 emotion classes (positive, negative and neutral).
• From this classifier, we labelised the all dataset WikiArt.
• With this new annoted dataset (more than 90 000 images), we created a Deep Conditional GAN (64*64) using pytorch, we trained it for 4 seperated style (portrait, abstract, landscape, flower-painting).
• For now thoses Gan's (generators) are store in ./weights folder.

Requirements

• torch
• numpy
• PIL
• matplotlib
• flask

Usage

git clone https://github.com/MarvinMartin24/CGAN-Emotion-Art/
# STYLE can be portrait or abstract or landscape or flower-painting
# EMOTION can be positive or negative or neutral
# s will save the images in animation/img
python3 generator_64.py STYLE EMOTION NUM_IMG_TO_GENERATE s

Example

python3 generator_64.py landscape positive 20 s
python3 generator_64.py portrait negative 2