An Improved EfficientNetV2 Model Based on Visual Attention Mechanism: Application to Identification of Cassava Disease

Table 9

Different network performance results on augmentation dataset. Infer time is measured on 1080ti GPU with batch size 16 using the same codebase; train time is the total training time. All models are trained with transfer learning.

Model	Top1 Acc (%)	Param (M)	FLOPs (G)	Infer time (s)	Train time (h)
AlexNet	89.8	57.0	0.8	2.01	4.8
VGG16	96.3	102	15.5	1.58	22.4
GoogLeNet	94.9	10.2	1.52	1.02	7.8
ResNet34	97.8	22.3	3.68	1.306	8.7
RegNetX	95.6	5.5	0.43	1.21	25
RegNetY	98.4	5.1	0.42	1.32	27
EfficientNetV2	98.53	21.4	2.908	1.01	35
PDRNet	99.56	21.5	2.909	1.01	35