|
Figure 2
Rare event detection workflow with three phases. The first phase (orange dashed rectangle) trains an image representation model (trained encoder) using a baseline data set for feature extraction. The trained encoder from the first phase is applied to a reference data set followed by the K-means clustering algorithm to obtain K centers to characterize the reference data set in the second phase (blue dashed rectangle). The output of the trained encoder from the first phase and the clustering model from the second phase, applied on the testing data set, is thresholded to determine REI for the testing data set in the third phase (green dashed rectangle). The different hyperparameters at each step are shown in red text. |
Open
access
![Open Access](/logos/open.png)