Using AAEHS-Net as an Attention-Based Auxiliary Extraction and Hybrid Subsampled Network for Semantic Segmentation

Table 1

Details of CEEM.

Blocks	Input size	Layers	Output size
CEEM1	128 × 128 64 channels	MSM1 ADM1	128 × 128 64 channels
MSM1	128 × 128 64 channels	(Doule conv k = 3, c = 64, s = 1) × 3	128 × 128 64 channels
ASM1	128 × 128 64 channels	Conv2D (k = 1, c = 64, s = 1)	128 × 128 64 channels
		Dilated conv (k = 3, c = 64, r = 4)
CEEM2	64 × 64 128 channels	MSM2 ADM2	64 × 64 128 channels
MSM2	64 × 64 128 channels	(Doule conv k = 3, c = 128, s = 1) × 4	64 × 64 128 channels
ASM2	64 × 64 128 channels	Conv2D (k = 1, c = 128, s = 1)	64 × 64 128 channels
		Dilated conv (k = 3, c = 128, r = 3)
CEEM3	32 × 32 256 channels	MSM3 ADM3	32 × 32 256 channels
MSM3	32 × 32 256 channels	(Doule conv k = 3, c = 256, s = 1) × 5	32 × 32 256 channels
ASM3	32 × 32 256 channels	Conv2D (k = 1, c = 256, s = 1)	32 × 32 256 channels
		Dilated conv (k = 3, c = 256, r = 2)
CEEM4	16 × 16 512 channels	MSM4	16 × 16 512 channels
MSM4	16 × 16 512 channels	(Doule conv k = 3, c = 512, s = 1) × 2	16 × 16 512 channels