Table 2.

Five CapsNet-based models with different parameters in the first step.

Input Data	Model	Conv. with ReLU	Primary Capsule	EmotionCaps	CapsNet Output
(18 × 18) × 1	1	3 × 3 × 32	3 × 3× (1 × 32)	(7 × 7) × 32	2 × 16
	2	3 × 3 × 128	3 × 3 × (1 × 128)	(7 × 7) × 128	2 × 16
	3	3 × 3 × 256	3 × 3 × (1 × 256)	(7 × 7) × 256	2 × 16
	4	3 × 3 × 384	3 × 3 × (1 × 384)	(7 × 7) × 384	2 × 16
	5	3 × 3 × 512	3 × 3 × (1 × 512)	(7 × 7) × 512	2 × 16