TensorFlow code for AnoVAEGAN with support for multiple GPU computation on 1 system.
This code is based on Deep Autoencoding Models for Unsupervised Anomaly Segmentation in Brain MR Images Paper. (Ref: https://arxiv.org/abs/1804.04488)
I have attempted to optimize the code for use on multi-GPU HPC using methods like tensorflow.distribute.MirroredStrategy(), tensorflow.keras.mixedprecision and some other methods based on my study about code optimization from various sources.
The code will work best for NVIDIA GPUs with compute capability 7.0 or higher like RTX, V100 and A100 GPUs. Other GPU systems could experience a decreased performance due to non-availability of Tensor Cores.