Domain Adaptation through Photorealistic Enhanced Images for Semantic Segmentation

Table 1

Semantic segmentation results of adapting GTAV to Cityscapes.

Methods	Road	Sidewalk	Building	Wall	Fence	Pole	Light	Sign	Veg	Terrain
Baseline (ResNet)	75.8	16.8	77.2	12.5	21.0	25.5	30.1	20.1	81.3	24.6
AdvEnt [8]	89.9	36.5	81.6	29.2	25.2	28.5	32.3	22.4	83.9	34.0
AdaSegNet [19]	86.5	36.0	79.9	23.4	23.3	23.9	35.2	14.8	83.4	33.3
CLAN [7]	87.0	27.1	79.6	27.3	23.3	28.3	35.5	24.2	83.6	27.4
Ours	89.4	46.0	83.1	27.6	22.7	33.6	33.6	27.3	83.6	34.5
IntraDA [9]	90.6	36.1	82.6	29.5	21.3	27.6	31.4	23.1	85.2	39.3
Ours + Intra	91.9	49.0	84.2	29.2	24.7	33.0	34.0	34.9	84.6	39.4
Methods	Sky	Person	Rider	Car	Truck	Bus	Train	Mbike	Bike	mIoU
Baseline (ResNet)	70.3	53.8	26.4	49.9	17.2	25.9	6.5	25.3	36.0	36.6
AdvEnt [8]	77.1	57.4	27.9	83.7	29.4	39.1	1.5	28.4	23.3	43.8
AdaSegNet [19]	75.6	58.5	27.6	73.7	32.5	35.4	3.9	30.1	28.1	42.4
CLAN [7]	74.2	58.6	28.0	76.2	33.1	36.7	6.7	31.9	31.4	43.2
Ours	78.1	59.4	29.8	79.6	36.5	41.6	0.1	23.6	25.3	45.0
IntraDA [9]	80.2	59.3	29.4	86.4	33.6	53.9	0.0	32.7	37.6	46.3
Ours + intra	81.4	59.8	29.8	84.2	35.3	44.9	0.0	28.8	33.7	47.5