Scene Understanding Based on High-Order Potentials and Generative Adversarial Networks

Table 1

Each class accuracy.

Class Label	FCN-32s	FCN-16s	FCN-8s	DeepLab	Our Method
Background	92.8	92.8	91.2	92.6	93.8
aeroplane	75.4	76.2	76.8	83.5	86.1
bicycle	33.6	34.3	34.4	36.6	35.9
bird	67.7	68.2	68.9	82.5	87.7
boat	48.6	49.4	49.4	62.3	63.5
bottle	58.4	59.2	60.3	66.5	67.2
bus	73.4	74.6	75.3	85.4	87.1
car	74.2	73.2	74.4	78.5	82.3
cat	77.6	78.4	77.6	73.7	86.8
chair	21.8	22.5	21.4	30.4	32.3
cow	62.1	62.5	62.5	72.9	76.5
Dining-table	46.3	46.7	46.8	60.4	62.0
dog	68.4	69.8	71.8	78.5	81.1
horse	63.4	63.8	63.9	75.5	77.9
motorbike	76.2	76.4	76.5	82.1	84.3
person	72.3	72.4	73.9	79.7	82.4
Potted-plant	44.5	44.5	45.2	58.2	59.6
sheep	71.2	71.6	72.4	82.0	84.3
sofa	37.4	37.2	37.4	48.8	54.9
train	69.4	69.8	70.9	73.7	76.2
tv/monitor	54.3	54.5	55.1	63.3	64.2