From: High-throughput soybean seeds phenotyping with convolutional neural networks and transfer learning
Dataset | Large dataset in the image size of 256 × 256 | |||||||
---|---|---|---|---|---|---|---|---|
Pre-trained model | Pre-trained COCO weights [35] | |||||||
Backbone layer | ResNet50-FPN | ResNet101-FPN | ||||||
Test dataset | Synthetic | Real-world | Synthetic | Real-world | ||||
256 | 512 | 1024 | 256 | 512 | 1024 | |||
Recall50 | 0.99 | 0.99 | 0.97 | 0.86 | 0.99 | 1.0 | 0.97 | 1.0 |
AP50 | 0.99 | 0.99 | 0.98 | 0.83 | 0.99 | 1.0 | 0.98 | 1.0 |
AP75 | 0.99 | 0.93 | 0.95 | 0.64 | 0.99 | 1.0 | 0.98 | 0.98 |
AP@[0.5:0.95] | 0.78 | 0.68 | 0.65 | 0.50 | 0.90 | 0.85 | 0.82 | 0.72 |