Detection of anomalous machine parts using deep convolution auto-encoders
- Add images to data/inputs folder.
- Train keras-retinanet for the above images.
- Run detection_image.py to generate cropped images for training anomaly detector in data/train
- Run generate_data.py to generate augmented data for train and valid set.
-
Deep Convolutional Model: (better results) -> Train model by running train.py -> test on images in data/test by running test.py. Anomalous images are saved in data/anomaly.
-
VGG based Convolutional Model: -> Train model by running vgg_train.py -> test on images in data/test by running vgg_test.py. Anomalous images are saved in data/anomaly.