AdvU-Net: Generating Adversarial Example Based on Medical Image and Targeting U-Net Model

Table 1

Architecture of the target model.

	Level	Conv layer	Filter	Stride	Output size
Input					244, 244, 3
Contracting path	Level 1	Conv 1	3, 3, 64	1	224, 244, 64
		Conv 2	3, 3, 64	1	224, 244, 64
	Level 2	Conv 3	3, 3, 128	2	112, 112, 128
		Conv 4	3, 3, 128	1	112, 112, 128
	Level 3	Conv 5	3, 3, 256	2	56, 56, 256
		Conv 6	3, 3, 256	1	56, 56, 256
	Level 4	Conv 7	3, 3, 512	2	28, 28, 512
		Conv 8	3, 3, 512	1	28, 28, 512
Bridge	Level 5	Conv 9	3, 3, 1024	2	14, 14, 1024
		Conv 10	3, 3, 1024	1	14, 14, 1024
Expansive path	Level 6	Conv 11	3, 3, 512	1	28, 28, 512
		Conv 12	3, 3, 512	1	28, 28, 512
	Level 7	Conv 13	3, 3, 256	1	56, 56, 256
		Conv 14	3, 3, 256	1	56, 56, 256
	Level 8	Conv 15	3, 3, 128	1	112, 112, 128
		Conv 16	3, 3, 128	1	112, 112, 128
	Level 9	Conv 17	3, 3, 64	1	224, 244, 64
		Conv 18	3, 3, 64	1	224, 244, 64
Output		Conv 19	1, 1	1	224, 244, 1