Table 4 Performance of custom YOLOv5, v7 and v8 models on FICS PCB standard dataset.
From: Advancing e-waste classification with customizable YOLO based deep learning models
Models | Class | Images | Instances | Precision | Recall | mAP50 | mAP50-95 |
|---|---|---|---|---|---|---|---|
YOLOv5 | All | 10 | 501 | 0.52 | 0.236 | 0.207 | 0.126 |
0 | 10 | 65 | 0.45 | 0.892 | 0.892 | 0.6 | |
1 | 10 | 231 | 0.287 | 0.377 | 0.377 | 0.0921 | |
2 | 10 | 174 | 0.381 | 0.145 | 0.145 | 0.0608 | |
3 | 10 | 3 | 0 | 0 | 0 | 0.00469 | |
4 | 10 | 3 | 1 | 0 | 0 | 0 | |
5 | 10 | 25 | 1 | 0 | 0 | 0 | |
Time: 100 epochs, 0.100 h Model summary (fused): 212 layers, 20,877,180 parameters, 0 gradients, 47.9 GFLOPs | |||||||
YOLOv7 | All | 10 | 501 | 0.187 | 0.0667 | 0.00453 | 0.00100 |
0 | 10 | 65 | 0.0692 | 0.108 | 0.0136 | 0.00426 | |
1 | 10 | 231 | 0.0223 | 0.177 | 0.00665 | 0.00102 | |
2 | 10 | 174 | 0.0332 | 0.115 | 0.00691 | 0.00126 | |
3 | 10 | 3 | 0 | 0 | 0 | 0 | |
4 | 10 | 3 | 1 | 0 | 0 | 0 | |
5 | 10 | 25 | 0 | 0 | 0 | 0 | |
Time: 100 epochs, 0.540 h Model Summary: 1032 layers, 151,055,568 parameters, 151,055,568 gradients, 210.0 GFLOPS | |||||||
YOLOv8 | All | 10 | 501 | 0.495 | 0.395 | 0.437 | 0.316 |
0 | 8 | 65 | 0.682 | 0.923 | 0.863 | 0.792 | |
1 | 8 | 231 | 0.573 | 0.459 | 0.461 | 0.277 | |
2 | 8 | 174 | 0.738 | 0.454 | 0.627 | 0.376 | |
3 | 2 | 3 | 0.143 | 0.333 | 0.159 | 0.143 | |
4 | 1 | 3 | 0 | 0 | 0 | 0 | |
5 | 3 | 25 | 0.833 | 0.2 | 0.512 | 0.306 | |
Time: 100 epochs, 0.535 h Model summary (fused): 218 layers, 25,843,813 parameters, 0 gradients Speed: 0.5 ms preprocess, 9.7 ms inference, 0.0 ms loss, 1.1 ms post process per image | |||||||
