Research Article
Object Detection Algorithm Based on Improved Feature Pyramid
Table 1
PASCAL VOC 2007 test detection results.
| Two-stage | One-stage | Method | Backbone | mAP (%) | FPS | Method | Backbone | mAP (%) | FPS |
| Faster R-CNN | VGG-16 | 73.2 | 7 | FA-SSD [24] | ResNet50 | 78.1 | 30 | Faster R-CNN | ResNet-101 | 76.4 | 2.4 | EFIPNet512 [25] | VGG-16 | 81.8 | 28 | ION | VGG-16 | 76.5 | 1.25 | SSD512 | VGG-16 | 79.5 | 19 | MR-CNN | VGG-16 | 78.2 | 0.03 | DSSD321 | ResNet-101 | 78.6 | 9.5 | Faster R-CNNā | ResNet-50-FPN | 79.4 | 19.1 | DSSD512 | ResNet-101 | 81.5 | 5.5 | Faster R-CNNā | ResNet-101-FPN | 81.6 | 16.0 | DFPR512 [26] | VGG-16 | 81.1 | 34 | R-FCN | ResNet-101 | 80.5 | 9 | RefineDet320 [27] | VGG-16 | 80.0 | 40.3 | CoupleNet[23] | ResNet-101 | 82.7 | 8.2 | RefineDet512 | VGG-16 | 81.8 | 24.1 | Ours | ResNet-50-AFPN | 82.2 | 16.8 | RFB Net 300 | VGG-16 | 80.5 | 83 | Ours | ResNet-101-AFPN | 83.5 | 14.1 | RFB Net 512 | VGG-16 | 82.2 | 38 |
|
|