Table 9 Comparison of PFLO with state-of-the-art pose estimation 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) |
|---|---|---|---|---|---|---|---|
HigherHRNet | HRNet-w48 | Bottom-up | 41.6 | 39.0 | 42.1*** | 14.3*** | 154 |
DEKR | Pose_HRNet_w48 | Bottom-up | 41.0 | 22.7 | 38.5*** | 34.9*** | 973 |
YOLOv5x | CSPDarknet | Top-down | 71.4 | 66.2 | 66.9*** | 36.7*** | 18.3 |
YOLOv6x | CSPDarknet | Top-down | 67.2 | 62.8 | 61.5*** | 32.1*** | 17.5 |
YOLOv8x | CSPDarknet | Top-down | 71.3 | 66.0 | 68.1** | 38.3* | 18.6 |
YOLOv9e | CSPDarknet | Top-down | 72.8 | 68.9 | 69.9*** | 39.8** | 16.7 |
YOLOv10x | CSPDarknet | Top-down | 73.3 | 67.8 | 69.5* | 38.5** | 23.7 |
YOLO11x | CSPDarknet | Top-down | 73.8 | 67.5 | 69.4*** | 38.5*** | 15.9 |
PFLO | CSPDarknet | Top-down | 75.2 ± 0.4 | 70.2 ± 1.2 | 72.2 ± 2.6 | 42.7 ± 3.6 | 24.7 |