Features to Text: A Comprehensive Survey of Deep Learning on Semantic Segmentation and Image Captioning

Table 1

Class pixel label distribution in the CamVid dataset.

Dataset	Method	mIoU
CamVid	ApesNet [90]	48.0
	ENet [91]	51.3
	SegNet [60]	55.6
	LinkNet [92]	55.8
	FCN8 [59]	57.0
	AttentionM [93]	60.1
	DeepLab-LFOV [72]	61.6
	Dilation8 [66]	65.3
	BiseNet [94]	68.7
	PSPNet [60]	69.1
	DenseDecoder [67]	70.9
	AGNet [95]	75.2
PASCAL VOC	Wails [96]	55.9
	FCN8 [59]	62.2
	PSP-CRF [97]	65.4
	Zoom Out [98]	69.6
	DCU [99]	71.7
	DeepLab1 [72]	71.6
	DeConvNet [61]	72.5
	GCRF [100]	73.2
	DPN [101]	74.1
	Piecewise [102]	75.3
Cityscapes	FCN8 [59]	65.3
	DPN [101]	66.8
	Dilation10 [103]	67.1
	LRR [104]	69.7
	DeepLab2 [73]	70.4
	FRRN [105]	71.8
	RefineNet [106]	73.6
	GEM [107]	73.69
	PEARL [108]	75.4
	TuSimple [109]	77.6
	PSPNet [110]	78.4
	SPP-DCU [99]	78.9