Open Santana18 opened 6 years ago
one way i think can try is to encode ur image into a fixed-length vector (that is feature extraction) using a pretrained model such as VGG or inception resnet, then feed them into the anomaly detection for training and outlier detection.
see topic. Thanks.