Research on Pedestrian Detection Algorithm Based on MobileNet-YoLo

Table 3

MobileNetv3 structural parameters.

Input^1	Operator^2	Exp size^3	Out^4	SE^5	NL^6	S^7
	Conv2d	—	16	—	HS	2
	Bottleneck, 3 × 3	16	16	—	RE	1
	Bottleneck, 3 × 3	64	24	—	RE	2
	Bottleneck, 3 × 3	72	24	—	RE	1
	Bottleneck, 5 × 5	72	40	✔	RE	2
	Bottleneck, 5 × 5	120	40	✔	RE	1
	Bottleneck, 5 × 5	120	40	✔	RE	1
	Bottleneck, 3 × 3	240	80	—	HS	2
	Bottleneck, 3 × 3	200	80	—	HS	1
	Bottleneck, 3 × 3	184	80	—	HS	1
	Bottleneck, 3 × 3	184	80	—	HS	1
	Bottleneck, 3 × 3	480	112	✔	HS	1
	Bottleneck, 3 × 3	672	112	✔	HS	1
	Bottleneck, 5 × 5	672	160	✔	HS	2
	Bottleneck, 5 × 5	960	160	✔	HS	1
	Bottleneck, 5 × 5	960	160	—	HS	1

¹Input represents the shape change of each feature layer; ²Operator represents the block structure that each feature layer is about to experience; ³Exp size, ⁴Out represent the number of channels that rise in the inverse residual structure within the neck, and the number of channels in the feature layer at the time of input to the neck, respectively; ⁵SE represents whether the attention mechanism is introduced at this layer; ⁶NL represents the type of activation function, HS represents h-swish, and RE represents RELU; ⁷S represents the step length used for each block structure.