Open Access
Table 3
Overview of object detection models used in the study.
| Model | Backbone | Layers | Parameters | Speed (FPS) | Pretrained dataset |
|---|---|---|---|---|---|
| SSD | MobileNet v2 | 324 | 4627367 | 4.55 | COCO 2017 |
| Faster R-CNN | ResNet50 V1 | 275 | 28278415 | 14.29 | COCO 2017 |
| YOLOv5 | CSP-Darknet | 270 | 8923647 | 100 | COCO 2017 |
| EfficientDet D1 | EfficientNet B1 | 849 | 4991810 | 1.72 | COCO 2017 |
| Non-pretrained CNN | custom Conv2D layers(a) | 21 | 810523 | 11.11 | – |
Notes. (a)See Sect. 4.5. Layers refer to the total number of layers and parameters are the total number of trainable parameters. Speed corresponds to detection speed in frame per second. The values differ from the defaults as the models were modified to fit our task.
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.