Deep Transfer Learning for Biology Cross-Domain Image Classification

Table 7

Results of cross-domain transfer learning on QUT Fish.

		Transfer process			Model	Accuracy (%)
		Flowers17		QUT Fish	GoogLeNet-v3	38.99	0.3093
		Flowers102		QUT Fish	GoogLeNet-v3	37.10	0.2899
		Plant Seedlings		QUT Fish	GoogLeNet-v3	39.36	0.3110
		PlanktonSet 1.0		QUT Fish	GoogLeNet-v3	50.33	0.4139
ImageNet		Flowers17		QUT Fish	GoogLeNet-v3	57.74	0.4725
ImageNet		Flowers102		QUT Fish	GoogLeNet-v3	58.09	0.4828
ImageNet		Plant Seedlings		QUT Fish	GoogLeNet-v3	54.17	0.4451
ImageNet		PlanktonSet 1.0		QUT Fish	GoogLeNet-v3	53.51	0.4515
		Flowers17		QUT Fish	ResNet-18	40.77	0.3178
		Flowers102		QUT Fish	ResNet-18	36.34	0.2807
		Plant Seedlings		QUT Fish	ResNet-18	38.08	0.2974
		PlanktonSet 1.0		QUT Fish	ResNet-18	45.46	0.3637
ImageNet		Flowers17		QUT Fish	ResNet-18	51.99	0.4191
ImageNet		Flowers102		QUT Fish	ResNet-18	51.	0.
ImageNet		Plant Seedlings		QUT Fish	ResNet-18	48.99	0.3951
ImageNet		PlanktonSet 1.0		QUT Fish	ResNet-18	49.42	0.4051
		Flowers17		QUT Fish	ResNet-34	40.93	0.3223
		Flowers102		QUT Fish	ResNet-34	37.16	0.2882
		Plant Seedlings		QUT Fish	ResNet-34	38.92	0.3004
		PlanktonSet 1.0		QUT Fish	ResNet-34	46.86	0.3782
ImageNet		Flowers17		QUT Fish	ResNet-34	52.65	0.4220
ImageNet		Flowers102		QUT Fish	ResNet-34	52.25	0.4235
ImageNet		Plant Seedlings		QUT Fish	ResNet-34	49.39	0.3956
ImageNet		PlanktonSet 1.0		QUT Fish	ResNet-34	49.54	0.4012
		Flowers17		QUT Fish	ResNet-50	35.74	0.2793
		Flowers102		QUT Fish	ResNet-50	31.61	0.2443
		Plant Seedlings		QUT Fish	ResNet-50	35.77	0.2721
		PlanktonSet 1.0		QUT Fish	ResNet-50	39.98	0.3151
ImageNet		Flowers17		QUT Fish	ResNet-50	52.34	0.4242
ImageNet		Flowers102		QUT Fish	ResNet-50	51.96	0.4222
ImageNet		Plant Seedlings		QUT Fish	ResNet-50	49.22	0.3979
ImageNet		PlanktonSet 1.0		QUT Fish	ResNet-50	43.25	0.3477

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