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

Table 6

Training parameters using RMS Prop in the model.

CNN layer	Operation	Output	Stride	Kernel	In	Out	Trainable variables
64 × 64 × 32	Convolution	32 × 32 × 32	2	2	32	32	0
32 × 32 × 32		32 × 32 × 64	2	3	32	64	1352
32 × 32 × 64	Max pooling	16 × 16 × 64	2	3	64	64	0
16 × 16 × 64	Conv	16 × 16 × 128	2	3	64	128	4256
16 × 16 × 128	Conv	8 × 8 × 128	2	3	128	128	45897
8 × 8 × 128	Max pooling	8 × 8 × 256	2	3	128	256	0
8 × 8 × 256	Conv	16 × 16 × 128	2	3	256	128	75686
16 × 16 × 128	Conv	16 × 16 × 256	2	3	128	256	77568
16 × 16 × 256	Conv	16 × 16 × 128	2	3	256	128	145689
16 × 16 × 128	Max pooling	32 × 32 × 64	2	3	128	64	0
32 × 32 × 64	Conv	32 × 32 × 128	2	3	64	128	214563
32 × 32 × 128	Conv	32 × 32 × 64	2	3	128	64	458796
32 × 32 × 64	Conv	64 × 64 × 32	2	3	64	32	8458923
64 × 64 × 32	Max pooling	64 × 64 × 64	2	3	32	64	0
64 × 64 × 64	Conv	64 × 64 × 32	2	3	63	32	7589666
128 × 128 × 16	Conv	128 × 128 × 32	2	3	16	32	47996669
128 × 128 × 32	Conv	128 × 128 × 16	2	3	32	16	47987642
14 × 14 × 512	Flatten	7 × 7 × 512	1	3	512	512	75852252
1 × 1 × 4096	FC2	1 × 1 × 1000	7	2	4096	1000	14256335