Deep Transfer Learning for Biology Cross-Domain Image Classification

Table 11

Results of cross-domain transfer learning on Flowers102.

		Transfer process			Model	Accuracy (%)
		Flowers17		Flowers102	GoogLeNet-v3	64.11	0.6228
		Plant Seedlings		Flowers102	GoogLeNet-v3	61.42	0.5948
		PlanktonSet 1.0		Flowers102	GoogLeNet-v3	73.95	0.7212
		QUT Fish		Flowers102	GoogLeNet-v3	61.98	0.5948
ImageNet		Flowers17		Flowers102	GoogLeNet-v3	91.19	0.9054
ImageNet		Plant Seedlings		Flowers102	GoogLeNet-v3	85.01	0.8407
ImageNet		PlanktonSet 1.0		Flowers102	GoogLeNet-v3	79.17	0.7788
ImageNet		QUT Fish		Flowers102	GoogLeNet-v3	86.21	0.8533
		Flowers17		Flowers102	ResNet-18	66.06	0.6421
		Plant Seedlings		Flowers102	ResNet-18	59.34	0.5675
		PlanktonSet 1.0		Flowers102	ResNet-18	72.39	0.7061
		QUT Fish		Flowers102	ResNet-18	61.81	0.5932
ImageNet		Flowers17		Flowers102	ResNet-18	87.88	0.8698
ImageNet		Plant Seedlings		Flowers102	ResNet-18	81.36	0.8074
ImageNet		PlanktonSet 1.0		Flowers102	ResNet-18	75.09	0.7308
ImageNet		QUT Fish		Flowers102	ResNet-18	79.92	0.7821
		Flowers17		Flowers102	ResNet-34	65.67	0.6400
		Plant Seedlings		Flowers102	ResNet-34	59.39	0.5773
		PlanktonSet 1.0		Flowers102	ResNet-34	72.30	0.7013
		QUT Fish		Flowers102	ResNet-34	63.34	0.6072
ImageNet		Flowers17		Flowers102	ResNet-34	88.50	0.8788
ImageNet		Plant Seedlings		Flowers102	ResNet-34	95.00	0.9496
ImageNet		PlanktonSet 1.0		Flowers102	ResNet-34	76.24	0.7477
ImageNet		QUT Fish		Flowers102	ResNet-34	80.84	0.7943
		Flowers17		Flowers102	ResNet-50	58.89	0.5688
		Plant Seedlings		Flowers102	ResNet-50	55.10	0.5324
		PlanktonSet 1.0		Flowers102	ResNet-50	61.70	0.5922
		QUT Fish		Flowers102	ResNet-50	59.10	0.5662
ImageNet		Flowers17		Flowers102	ResNet-50	89.28	0.8877
ImageNet		Plant Seedlings		Flowers102	ResNet-50	83.22	0.8199
ImageNet		PlanktonSet 1.0		Flowers102	ResNet-50	77.85	0.7600
ImageNet		QUT Fish		Flowers102	ResNet-50	84.68	0.8359

indicates the results outperform the corresponding results of training from scratch and fine-tuning on ImageNet.