Modifications | Model | Precision (%) | Recall rate (%) | F1-score (%) | mAP (%) | Inference time (ms) | B-FLOPS | Size (MB) |
---|---|---|---|---|---|---|---|---|
Original | YOLOv4 | 61 | 86 | 72 | 81.43 | 905.65 | 59.57 | 244.40 |
Residual Block pruning | YOLOv4-RC3 | 50 | 95 | 66 | 88.91 | 695.54 | 47.59 | 242.40 |
YOLOv4-RC4 | 50 | 91 | 65 | 85.20 | 695.98 | 51.21 | 233.20 | |
YOLOv4-RC5 | 61 | 93 | 73 | 89.84 | 731.43 | 57.61 | 222.10 | |
YOLOv4-RC3_4 | 59 | 96 | 74 | 90.70 | 684.93 | 37.35 | 221.50 | |
YOLOv4-RC3_5 | 59 | 93 | 72 | 88.09 | 690.53 | 45.64 | 220.40 | |
Backbone replacement | YOLOv4- ResNet-50L | 54 | 83 | 65 | 76.95 | 892.80 | 37.33 | 209.30 |
YOLOv4-ResNet-50Â M | 65 | 81 | 72 | 78.96 | 905.39 | 37.33 | 209.30 |