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

Table 6 The quantitative of evaluation metrics of the model retrained by small dataset of 256 × 256 image size with COCO weights

Dataset	Small 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
Test dataset	256	512	1024	Real-world	256	512	1024	Real-world
Recall₅₀	0.75	0.84	0.79	0.45	0.88	0.65	0.57	0.91
AP₅₀	0.76	0.85	0.82	0.54	0.92	0.65	0.62	0.91
AP₇₅	0.76	0.58	0.40	0.11	0.92	0.64	0.62	0.89
AP@[0.5:0.95]	0.67	0.51	0.43	0.22	0.80	0.50	0.47	0.66

ISSN: 1746-4811