Table 1.

Layer disposal of our proposed network. “Out-F”: Number of Feature maps at the layer’s Output, “Out-Res”: Output Resolution for input size of 640 × 480 RGB images from the smart glasses, “C”: number of semantic prediction Classes. Encoder: Layers 1–16, decoder: Layers 17–19.

Layer	Type	Out-F	Out-Res
0	Scaling 640 × 480	3	320 × 240
1	Down-sampler block	16	160 × 120
2	Down-sampler block	64	80 × 60
3–7	5 × Non-bt-1D	64	80 × 60
8	Down-sampler block	128	40 × 30
9	Non-bt-1D (dilated 2)	128	40 × 30
10	Non-bt-1D (dilated 4)	128	40 × 30
11	Non-bt-1D (dilated 8)	128	40 × 30
12	Non-bt-1D (dilated 16)	128	40 × 30
13	Non-bt-1D (dilated 2)	128	40 × 30
14	Non-bt-1D (dilated 4)	128	40 × 30
15	Non-bt-1D (dilated 8)	128	40 × 30
16	Non-bt-1D (dilated 2)	128	40 × 30
17a	Original feature map	128	40 × 30
17b	Pooling and convolution	32	40 × 30
17c	Pooling and convolution	32	20 × 15
17d	Pooling and convolution	32	10 × 8
17e	Pooling and convolution	32	5 × 4
17	Up-sampler and concatenation	256	40 × 30
18	Convolution	C	40 × 30
19	Up-sampler	C	640 × 480