PF-ViT: Parallel and Fast Vision Transformer for Offline Handwritten Chinese Character Recognition

Table 4

Performance of different models on the DHWDB dataset: parameters; FLOPs; accuracy.

Methods	Number of encoder layers per channel	Epochs	#Params (M)	FLOPs (G)	Acc. (%)
T-ViT	3	300	43.11	4.32	98.1
	4	300	57.28	5.72	98.3
	6	300	85.62	8.52	98.6
F-ViT	2	300	57.28	2.94	96.6
	3	300	85.62	4.36	97.3
	6	300	170.63	8.61	97.7
S-ViT	2	300	99.79	2.99	96.3
	3	300	148.38	4.43	97.1
	4	300	198.98	5.86	97.0