A Real-Time Image Semantic Segmentation Method Based on Multilabel Classification

Table 4

Comparison of the experiment results on self-built datasets by category.

Network model	Buildings	Trees	Sky	Cars	Road	Pedestrians	Cyclists	MIoU (%)	Accuracy (%)
The proposed model	81.1	85.9	82.0	83.7	95.2	42.6	50.3	74.4	89.1
SegNet	78.8	79.9	90.5	83.2	92.9	51.2	40.8	73.9	88.8
RefineNet	78.2	74.5	92.1	83.5	93.5	46.5	48.1	73.8	88.4
DeepLab	81.5	73.2	88.3	84.1	94.9	48.4	50.1	74.4	89.1
PSPNet	80	74.2	90.5	83.1	90.5	49.7	47.3	73.6	88.3
LRR	76.4	79.3	92.5	81.5	89.7	55.2	47.5	74.6	89.2
Dilation-8	77.9	85.0	92.2	79.3	92.4	52.4	48.9	75.4	91.5
FCN-8s	79.9	81.5	88.6	76.8	92.2	41.6	40	71.5	84.3