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

Table 2

Details of algorithm steps.

Blocks	Input size	Output size
Input	512 × 512 × 3
Conv2d c = 64, k = 7, s = 2	512 × 512 × 3	256 × 256 × 64
Max pool	256 × 256 × 64	128 × 128 × 64
CEEM1	128 × 128 × 64	128 × 128 × 64
HSM1	128 × 128 × 64	64 × 64 × 64
CEEM2	64 × 64 × 64	64 × 64 × 128
HSM2	64 × 64 × 128	32 × 32 × 128
CEEM3	32 × 32 × 128	32 × 32 × 256
HSM3	32 × 32 × 256	16 × 16 × 256
CEEM4	16 × 16 × 256	16 × 16 × 512
Transpose2d	16 × 16 × 512	32 × 32 × 256
Concat	32 × 32 × 256	32 × 32 × 512
(Conv2d, c = 256)  2	32 × 32 × 512	32 × 32 × 256
Transpose2d	32 × 32 × 256	64 × 64 × 128
Concat	64 × 64 × 128	64 × 64 × 256
(Conv2d, c = 128)  2	64 × 64 × 256	64 × 64 × 128
Transpose2d	64 × 64 × 128	128 × 128 × 64
Concat	128 × 128 × 64	128 × 128 × 128
(Conv2d, c = 64)  2	128 × 128 × 128	128 × 128 × 64
Transpose2d	128 × 128 × 64	256 × 256 × 32
Concat	256 × 256 × 32	256 × 256 × 96
(Conv2d, c = 32)  2	256 × 256 × 96	256 × 256 × 32
Transpose2d c = 32	256 × 256 × 32	512 × 512 × 32
Conv2d c = 1	512 × 512 × 32	512 × 512 × 1
Output		512 × 512 × 1