TensorFlow implementation of f-AnoGAN with MNIST dataset [1].
The base model WGAN is also implemented with TensorFlow.

The architecture of f-AnoGAN [1].

The logic for calculating anomaly score [1].

Graph of f-AnoGAN.

'Class-1' is defined as normal and the others are defined as abnormal.

The rear half of the graph represents the state of the training phase 2.

Term Real	Term Fake

Loss D (Discriminator)	Loss G (Generator)

z:2	z:2 (latent space walking)

z:64	z:128

The front half of the graph represents the state of the training phase 1.

Term izi	Term ziz	Loss E (Encoder)

Restoration result by f-AnoGAN.

Box plot with encoding loss of test procedure.

• Python 3.7.4
• Tensorflow 1.14.0
• Numpy 1.17.1
• Matplotlib 3.1.1
• Scikit Learn (sklearn) 0.21.3

[1] Schlegl, Thomas, et al (2019). f-AnoGAN: Fast unsupervised anomaly detection with generative adversarial networks. Medical image analysis 54 (2019): 30-44.