This Outlier/Anomaly detection started from this question. It explores the features extracted and then use Sklearn's One Class Support Vector Machine (OC-SVM) to gain decent results.
- Load the dataset onto the RAM for faster execution
- To extract features, we use the pre-trained Resnet50 weights.
- Normalize the input using Standard scalar
- Reduce the number of channels using PCA
- Use GridSearch in order to find the parameters for One Class SVM.
- Plot the Confusion Matrix to know the accuracy.