Deep Learning with Taxonomic Loss for Plant Identification

Table 2

Accuracies of eight state-of-the-art neural networks trained by cross-entropy and taxonomic loss on PlantCLEF 2015 dataset.

Neural network	Loss function	Accuracy (%)
		Family	Genus	Species
GoogLeNet [13]	CL	72.62	65.97	59.69
	TAX	74.95	68.07	61.06
ResNet-50 [23]	CL	77.48	71.59	65.07
	TAX	78.55	72.20	65.15
Inception-v3 [14]	CL	77.93	74.01	67.98
	TAX	80.31	74.46	67.42
Inception-ResNet-v2 [11]	CL	80.66	74.57	67.93
	TAX	83.36	76.85	70.38
MobileNet v2 [24]	CL	72.13	65.65	59.16
	TAX	74.18	67.52	60.42
ShuffleNet v2 [25]	CL	66.39	59.32	52.88
	TAX	68.80	61.45	54.18
DenSeNet-169 [26]	CL	78.57	73.00	66.76
	TAX	80.11	74.48	67.23
SENet-154 [27]	CL	81.25	76.81	70.08
	TAX	83.19	78.08	71.15