yunseokddi / deepsad_pytorch

DeepSAD pytorch version

1. Overview

This is implementation from Deep semi-supervised Anomaly detection paper to pytorch version

I use CIFAR10 dataaset

1. How to run?

• Check the parser's parameters in main.py
• python3 main.py

2. Environment

python version = 3.6.8 (Recommand)

torch                   1.7.1+cu110
torchaudio              0.7.2
torchsummary            1.5.1
torchvision             0.8.2+cu110
tqdm                    4.61.2
imageio                 2.9.0
scikit-learn            0.24.2
numpy                   1.16.6

3. Result

Model	train loss	test loss	AUROC
AUTOENCODER (epoch 50)	0.00807	0.010426	66.50%
Deep SVDD (epoch 100)	0.06367	0.061671	77.30%

Tensorboard result

4. Reference

https://arxiv.org/abs/1906.02694 [Official paper]

https://github.com/lukasruff/Deep-SAD-PyTorch [Official paper code]

https://ys-cs17.tistory.com/51 [My paper review blog]

deepsad_pytorch's Issues

AUC-ROC

Hello, When I run this code on my dataset, I often find that AUC-ROC is less than AUC-PR.
Is there a problem with my data set or is this a normal situation?
I would appreciate your reply :)