Seite - 165 - in Joint Austrian Computer Vision and Robotics Workshop 2020

Feature Bagging (FB) 1 2 3 4 5 6 7 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 73 1 7 19 Recall = 0.73 Precision = 0.95 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 auROC = 0.96 auPR = 0.93 ROC PR 5-Nearest Neighbors (KNN) 0 2 4 6 8 10 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 74 0 6 20 Recall = 0.77 Precision = 1.00 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 auROC = 0.97 auPR = 0.94 ROC PR Local Outlier Factor (LOF) 1 2 3 4 5 6 7 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 72 2 9 17 Recall = 0.65 Precision = 0.89 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 auROC = 0.96 auPR = 0.91 ROC PR Minimum Covariance Determinant (MCD) 0 50 100 150 200 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 72 2 12 14 Recall = 0.54 Precision = 0.88 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 auROC = 0.95 auPR = 0.87 ROC PR One-class SVM (OCSVM) 0 1 2 3 4 5 increasing outlier scores 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 72 2 4 22 Recall = 0.85 Precision = 0.92 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 auROC = 0.88 auPR = 0.89 ROC PR Figure 4: Selected outlier detectors and their performance on test set, i.e., the segmentation results of the algorithm. 165