MRFDCNet: Multireceptive Field Dense Connection Network for Real-Time Semantic Segmentation

Table 2

Performance comparison among different methods on the Cityscapes test set.

Model	Input size	Pretrain	GFLOPs (G)	Params (M)	FPS	mIoU
ESPNetv2	512 × 1024	ImageNet	5.85	1.25	65	66.2
BiSeNet	768 × 1536	ImageNet	14.80	5.8	67	68.4
ICNet [37]	1024 × 2048	ImageNet	28.30	26.5	35	69.5
FPENet [38]	512 × 1024	No	1.58	0.12	110	55.2
CGNet	512 × 1024	No	7.01	0.5	64	64.8
EDANet [39]	512 × 1024	No	8.95	0.68	102	67.3
ESPNet	512 × 1024	No	4.7	0.36	113	60.3
DABNet	512 × 1024	No	10.46	0.76	104.2	70.1
ERFNet	512 × 1024	No	26.86	2.1	49	68
MRFDCNet	512 × 1024	No	13.06	1.07	74	72.8