adVAE
adVAE copied to clipboard
Implementation of 'Self-Adversarial Variational Autoencoder with Gaussian Anomaly Prior Distribution for Anomaly Detection'
adVAE (Self-Adversarial Variational Autoencoder)
Implementation of 'Self-Adversarial Variational Autoencoder with Gaussian Anomaly Prior Distribution for Anomaly Detection' (adVAE) [Related repository].
The official implementation is provided by WangXuhongCN.
Architecture
![](https://github.com/YeongHyeon/adVAE/raw/master/figures/advae.png)
Simplified adVAE architecture.
Graph in TensorBoard
![](https://github.com/YeongHyeon/adVAE/raw/master/figures/graph.png)
Graph of adVAE.
Results
![](https://github.com/YeongHyeon/adVAE/raw/master/figures/restoring.png)
Restoration result by adVAE.
![](https://github.com/YeongHyeon/adVAE/raw/master/figures/test-box.png)
![](https://github.com/YeongHyeon/adVAE/raw/master/figures/histogram-test.png)
Box plot and histogram of restoration loss in test procedure.
![](https://github.com/YeongHyeon/adVAE/raw/master/figures/test-latent_z.png)
![](https://github.com/YeongHyeon/adVAE/raw/master/figures/test-latent_z_T.png)
Normal latent space, and transformed latent space (regared as abnormal).
Environment
- Python 3.7.4
- Tensorflow 1.14.0
- Numpy 1.17.1
- Matplotlib 3.1.1
- Scikit Learn (sklearn) 0.21.3
Reference
[1] Xuhong Wang, et al. (2019). Self-adversarial Variational Autoencoder with Gaussian Anomaly Prior Distribution for Anomaly Detection.. arXiv preprint arXiv:1903.00904.