Table 2
PASCAL and COCO evaluation metrics from detection on the validation set using different models.
| Metric | SSD | Faster R-CNN | YOLO | EfficientDet D1 | NPCNN |
|---|---|---|---|---|---|
| PASCAL | |||||
| APP (IoU=0.50) | 0.992 | 0.991 | 0.989 | 0.965 | 0.409 |
| APM (IoU=0.50) | 0.959 | 0.998 | 0.999 | 0.895 | 0.201 |
| mAP (IoU=0.50) | 0.975 | 0.994 | 0.993 | 0.929 | 0.305 |
| COCO | |||||
| AP (IoU=0.50:0.05:0.95) | 0.839 | 0.833 | 0.915 | 0.726 | 0.086 |
| AP (IoU=0.50) | 0.969 | 0.990 | 0.985 | 0.925 | 0.298 |
| AP (IoU=0.75) | 0.952 | 0.969 | 0.964 | 0.868 | 0.028 |
| AR (max=1) | 0.855 | 0.848 | 0.919 | 0.762 | 0.202 |
Notes. The abbreviations AP, AR, and mAP stand for average precision, average recall and mean average precision. The IoU thresholds are written in parentheses for APs where IoU=0.50:0.05:0.95 refer to an IoU interval between 0.50 and 0.95 with a 0.05 increment. The AR (max=1) corresponds to one detection per image value. The term NPCNN stands for the custom non-pretrained CNN-based model.
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.