High-throughput soybean seeds phenotyping with convolutional neural networks and transfer learning

Table 8 The quantitation of evaluation metrics of the model retrained by the small dataset of 256 × 256 px with synthetic barley weights

Dataset	Small dataset in the image size of 256 × 256
Pre-trained model	Pre-trained barley weights [40]
Backbone layer	ResNet50-FPN				ResNet101-FPN
Test dataset	Synthetic			Real-world	Synthetic			Real-world
Test dataset	256	512	1024	Real-world	256	512	1024	Real-world
Recall₅₀	0.30	0.05	0.10	0.02	0.99	1.0	0.88	1.0
AP₅₀	0.33	0.05	0.12	0.01	0.99	1.0	0.93	1.0
AP₇₅	0.32	0.04	0.06	0.00	0.99	1.0	0.93	0.99
AP@[0.5:0.95]	0.24	0.03	0.07	0.00	0.90	0.83	0.77	0.72

ISSN: 1746-4811