experimented with 14 different types of augmentation approaches. The output image is generated using the following techniques: vertical concatenation, horizontal concatenation, mixed concatenation, random 2x2, VH- mixup (vertical concatenation, horizontal concate- nation, and mixup), VH-BC+ (vertical concatena- tion, horizontal concatenation, and between-class), random square, random column interval, random row interval, random rows, random columns, ran- dom pixels, random elements, and noisy mixup.
From all these approached, VHmixup has the best performance.