Crowd Counting Based on Multiresolution Density Map and Parallel Dilated Convolution

Table 1

Detailed architecture of MDPDNet convolutional layer.

Layers	Backbone
Input	Image
1	VGG-16 layers 1–4
Module	PDM	MDM
2	Conv7 × 7 × 256	VGG-16 layers 5–7
3	Conv5 × 5 × 512	Conv1 × 1 × 1 (b1)	VGG-16 layers 8–10
4	Conv3-64-1-256; Conv3-64-2-256; Conv3-64-5-256	Upsampling-2	Conv1 × 1 × 1 (b2)	VGG-16 layers 11–13
5		Conv3 × 3 × 64	Conv3 × 3 × 64	Conv1 × 1 × 1 (b3)
6			Upsampling-4	Upsampling-4
7				Conv3 × 3 × 64
8	Upsampling-2			Upsampling-4
9	Conv3 × 3 × 128	Conv3 × 3 × 128
10	Upsampling-2
11	Conv1 × 1 × 1
Output	HD