Attribute conditional face generation with ACGAN in keras; GPU computation of FID and attribute metrics.
An in depth description of the work can be found here: project_report.pdf.
An online interactive version of the model on is available as a Huggingface Space.
Conditional results with 10 attribtes input at 218x178 resolution:
attribute_conditional_gan.ipynb
is a self contained notebook for the training of models with different image resolution and number of attributes .
evaluate_GAN.ipynb
keeps the code for evaluation of trained models and plotting images with relative attributes.
GAN_Attribute_Classifiers.ipynb
has the code for training the MobilenetV2 attribute classifiers used to evaluate the GAN models with Accuracy and F1 score.