yeonghyeon / conad-pytorch

Implementation of Consistency-based anomaly detection (ConAD)

[PyTorch] Consistency-based anomaly detection (ConAD)

Implementation of Consistency-based anomaly detection (ConAD) from paper 'Anomaly Detection With Multiple-Hypotheses Predictions' with MNIST dataset [TensorFlow Version].

Architecture

Simplified ConAD architecture.

Graph in TensorBoard

Graph of ConAD.

Results

Restoration result by CondAD.

Box plot and histogram of restoration loss in test procedure.

Latent space of each class.

Environment

• Python 3.7.4
• PyTorch 1.1.0
• Numpy 1.17.1
• Matplotlib 3.1.1
• Scikit Learn (sklearn) 0.21.3

Reference

[1] Duc Tam Nguyen, et al. (2018 arXiv, 2019 ICML). Anomaly Detection With Multiple-Hypotheses Predictions. ICML 2019.

conad-pytorch's Issues

so weird net architecture and loss_func