Table 11.

The MAP(%) of attacked submodels and the ensemble models trained on the different dataset and YOLOv3 model, under attacks from different methods on the first submodel

Dataset Attack methods BEM
DEM
Model1 Model2 Model3 Ensemble Model1 Model2 Model3 Ensemble
VOC Non-Attack 53.6 55.0 55.1 59.8 57.1 56.5 55.9 60.0
DPAttack 19.5 37.4 38.6 42.7 12.6 41.4 41.4 45.5
TOG 0.1 43.9 44.4 48.6 0.3 45.6 50.4 51.2
DAG 15.5 38.4 40.1 35.1 19.8 43.2 44.3 41.1

Non-Attack 60.0 59.9 58.9 60.5 55.9 53.4 56.7 57.8
COCO DPAttack 19.2 26.2 25.1 27.5 21.3 33.3 33.6 37.6
TOG 2.6 9.6 10.0 10.7 2.2 50.2 49.2 53.4
DAG 17.0 19.6 19.6 22.7 16.4 46.5 47.5 43.6

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.