# Kaggle AD Challenge: Unsupervised Image Anomaly Detection Competition
## Project Goals:
- Students compete at Kaggle, a predictive model and analysis contest platform, and have hands-on experience and acquire deep learning knowledge through solving real-world image anomaly detection problem. Competition will be done by performing anomaly segmentation task for one category of MVTec AD image dataset. Evaluate project by measuring segmentation pixelwise AUROC(area under a receiver operating characteristic curve) performance.

## Dataset and Experimental Environment:
- Dataset: MVTec AD is a dataset for benchmarking anomaly detection methods with a focus on industrial inspection. It contains over 5000 high-resolution images divided into fifteen different object and texture categories. Each category comprises a set of defect-free training images and a test set of images with various kinds of defects as well as images without defects. In this competition, we compete using only one category (capsule) of the MVtec AD dataset. There is no anomaly in the train dataset.
- Experimental Environment:  Kaggle InClass competition, IPython notebook. Pytorch. Code for data loader and evaluation function will be provided.