Table 10 Comparison of PFLO with state-of-the-art object detection models
From: PFLO: a high-throughput pose estimation model for field maize based on YOLO architecture
Models | Backbone | Approach | P (%) | R (%) | mAP50 (%) | mAP50 - 95 (%) | Inference time (ms) |
|---|---|---|---|---|---|---|---|
Faster R-CNN | ResNet101 + FPN | Two-stage | 76.1 | 80.5 | 76.1*** | 41.3*** | 89.6 |
YOLOv5x | CSPDarknet | Single-stage | 85.6 | 83.5 | 90.2* | 60.9*** | 18.3 |
YOLOv6x | CSPDarknet | Single-stage | 83.0 | 82.0 | 87.9*** | 58.3*** | 17.5 |
YOLOv8x | CSPDarknet | Single-stage | 85.5 | 84.3 | 90.4* | 62.0*** | 18.6 |
YOLOv9e | CSPDarknet | Single-stage | 85.9 | 85.1 | 90.7 | 63.6* | 16.7 |
YOLOv10x | CSPDarknet | Single-stage | 86.2 | 85.0 | 91.2 | 63.3* | 23.7 |
YOLO11x | CSPDarknet | Single-stage | 86.5 | 85.0 | 91.2 | 62.7** | 15.9 |
PFLO | CSPDarknet | Single-stage | 86.8 ± 0.5 | 86.2 ± 0.8 | 91.6 ± 0.5 | 66.0 ± 1.8 | 24.7 |