A New Image Classification Approach via Improved MobileNet Models with Local Receptive Field Expansion in Shallow Layers

Table 3

Classification accuracy rates (%) on Caltech-256 dataset.

Number of iterations	30000	35000	40000	45000	50000
SqueezeNet	41.48	43.06	43.39	43.58	44.03
MobileNets	64.48	64.58	64.55	64.67	64.52
Dense1-MobileNet	64.61	64.53	64.45	64.44	64.47
Dense2-MobileNet	65.62	65.67	65.84	65.78	65.79
Dilated1-MobileNet	65.77	65.74	65.87	65.90	65.87
Dilated2-MobileNet	66.10	66.06	65.94	65.84	65.94
Dilated3-MobileNet	64.97	64.9	64.87	65.19	65.16

We also validate our method on the Animals with Attributes (AwA) dataset [28]. The classification accuracy rates are shown in Table 4.