mb-29 / gan-anomaly-detection

Image-to-Image Translation in PyTorch

Anomaly detection

Anomaly detection using cycle-GAN.

Originial Repository

Image-to-Image Translation in PyTorch

Setup

Please install CycleGAN network and download the pix2pix maps dataset following the instructions provided in the original repository. The checkpoint files and weights we obtained after a training phase of 11 epochs and that we used for our project are available here.

Compute and save embeddings

python anomaly_detection/features.py --dataroot ./datasets/[Data folder] --name maps_cyclegan --model cycle_gan --gpu_ids -1

Naive tests

Run

python anomaly_detection/data_analysis.py

Anomaly detection

Please find our work in the Jupyter notebooks here.