Table 5 Model architectures and results (%) comparison on RPC dataset (Easy & Not Densely placed products).
From: Smart retail SKUs checkout using improved residual network
Model | Backbone | Feature fusion | mAP | AP.75 | Parameter size | GFLOPs |
|---|---|---|---|---|---|---|
RetinaNet | ResNet50 | FPN | 72.3 | 74.7 | 36.3M | 182 |
RetinaNet | ResNet101 | FPN | 72.1 | 76.2 | 55.4M | 212.5 |
Faster-RCNN | ResNet50 | FPN | 69.8 | 73.3 | 41.8M | 134.7 |
Faster-RCNN | ResNet101 | FPN | 71.8 | 74.4 | 59.8M | 165.2 |
Faster-RCNN | VGG16 | RPN | 70.2 | 74.1 | 44.9M | 209.4 |
VovNet | VovNet | N/A | 69.3 | 75.1 | 35.8M | 75.3 |
YOLO | CSPResNext50 | PANet | 72.8 | 75.4 | 65M | 72.4 |
The proposed model | ResNet50 | FPN | 73.7 | 78.1 | 29.4M | 73.1 |
