Deep Neural Network for Accurate Age Group Prediction through Pupil Using the Optimized UNet Model

Table 2

Training summary of UNet.

Layer (depth -index)	Output shape	Training parameters
Downsampling
Conv2d: 3–1	32, 512, 512	320
BatchNorm2d: 3–3	32, 512, 512	64
Conv2d: 3–4	32, 512, 512	9,248
Conv2d: 3–8	64, 256, 256	18,496
BatchNorm2d: 3–10	64, 256, 256	128
Conv2d: 3–11	64, 256, 256	36,928
Conv2d: 3–15	128, 128, 128	73,856
BatchNorm2d: 3–17	128, 128, 128	256
Conv2d: 3–18	128, 128, 128	147,584
BatchNorm2d: 3–20	128, 128, 128	256
Conv2d: 3–22	256, 64, 64	295,168
BatchNorm2d: 3–24	256, 64, 64	512
Conv2d: 3–25	256, 64, 64	590,080
BatchNorm2d: 3–27	256, 64, 64	512
Upsampling
ConvTranspose2d: 3–28	128, 128, 128	131,200
BatchNorm2d: 3–30	128, 128, 128	256
Conv2d: 3–32	128, 128, 128	295,040
BatchNorm2d: 3–34	128, 128, 128	256
Conv2d: 3–35	128, 128, 128	147,584
BatchNorm2d: 3–37	128, 128, 128	256
ConvTranspose2d: 3–38	64, 256, 256	32,832
BatchNorm2d: 3–40	64, 256, 256	128
Conv2d: 3–42	64, 256, 256	73,792
BatchNorm2d: 3–44	64, 256, 256	128
Conv2d: 3–45	64, 256, 256	36,928
BatchNorm2d: 3–47	64, 256, 256	128
ConvTranspose2d: 3–48	32, 512, 512	8,224
BatchNorm2d: 3–50	32, 512, 512	64
Conv2d: 3–52	32, 512, 512	18,464
BatchNorm2d: 3–54	32, 512, 512	64
Conv2d: 3–55	32, 512, 512	9,248
BatchNorm2d: 3–57	32, 512, 512	64