Facial Expression Recognition Method Combined with Attention Mechanism

Table 1

Network setting of ERFNet.

Arrangement	Modular	Output	Feature graph size/pixel
1	Downsampler block	16	512 × 256
2	Downsampler block	64	256 × 128
3∼7	ARB × 5	64	256 × 128
8	Downsampler block	256	128 × 64
9	Non-bt-1D (dilated 2)	256	128 × 64
10	Non-bt-1D (dilated 4)	256	128 × 64
11	Non-bt-1D (dilated 8)	256	128 × 64
12	Non-bt-1D (dilated 16)	256	128 × 64
13	Non-bt-1D (dilated 2)	128	128 × 64
14	Non-bt-1D (dilated 4)	128	128 × 64
15	Non-bt-1D (dilated 8)	128	128 × 64
16	Non-bt-1D (dilated 16)	128	128 × 64
17	Upsampler block	64	256 × 128
18∼19	Non-bt-1D × 2	64	256 × 128
20	Upsampler block	16	512 × 256
21∼22	ARB × 2	16	512 × 256
23	Output	C	1024 × 512