Deep Transfer Learning for Biology Cross-Domain Image Classification

Table 6

Results of cross-domain transfer learning on Flowers17.

		Transfer process			Model	Accuracy (%)
		Flowers102		Flowers17	GoogLeNet-v3	90.00	0.9002
		Plant Seedlings		Flowers17	GoogLeNet-v3	88.529 4	0.8832
		PlanktonSet 1.0		Flowers17	GoogLeNet-v3	92.06	0.9198
		QUT Fish		Flowers17	GoogLeNet-v3	89.12	0.8912
ImageNet		Flowers102		Flowers17	GoogLeNet-v3	97.94	0.9795
ImageNet		Plant Seedlings		Flowers17	GoogLeNet-v3	95.88	0.9589
ImageNet		PlanktonSet 1.0		Flowers17	GoogLeNet-v3	96.47	0.9648
ImageNet		QUT Fish		Flowers17	GoogLeNet-v3	95.00	0.9500
		Flowers102		Flowers17	ResNet-18	88.82	0.8908
		Plant Seedlings		Flowers17	ResNet-18	90.88	0.9077
		PlanktonSet 1.0		Flowers17	ResNet-18	92.06	0.9198
		QUT Fish		Flowers17	ResNet-18	89.12	0.8912
ImageNet		Flowers102		Flowers17	ResNet-18	97.06	0.9703
ImageNet		Plant Seedlings		Flowers17	ResNet-18	93.82	0.9381
ImageNet		PlanktonSet 1.0		Flowers17	ResNet-18	93.53	0.9356
ImageNet		QUT Fish		Flowers17	ResNet-18	94.41	0.9436
		Flowers102		Flowers17	ResNet-34	90.88	0.9088
		Plant Seedlings		Flowers17	ResNet-34	89.41	0.8937
		PlanktonSet 1.0		Flowers17	ResNet-34	92.65	0.9265
		QUT Fish		Flowers17	ResNet-34	88.82	0.8884
ImageNet		Flowers102		Flowers17	ResNet-34	97.65	0.9763
ImageNet		Plant Seedlings		Flowers17	ResNet-34	95.00	0.9496
ImageNet		PlanktonSet 1.0		Flowers17	ResNet-34	93.53	0.9353
ImageNet		QUT Fish		Flowers17	ResNet-34	94.41	0.9441
		Flowers102		Flowers17	ResNet-50	84.41	0.8456
		Plant Seedlings		Flowers17	ResNet-50	89.41	0.8927
		PlanktonSet 1.0		Flowers17	ResNet-50	92.06	0.9203
		QUT Fish		Flowers17	ResNet-50	87.06	0.8689
ImageNet		Flowers102		Flowers17	ResNet-50	96.47	0.9650
ImageNet		Plant Seedlings		Flowers17	ResNet-50	96.18	0.9614
ImageNet		PlanktonSet 1.0		Flowers17	ResNet-50	94.12	0.9405
ImageNet		QUT Fish		Flowers17	ResNet-50	96.76	0.9675

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