Semantic Segmentation under a Complex Background for Machine Vision Detection Based on Modified UPerNet with Component Analysis Modules

Table 1

Pixel accuracy and segmentation time of the main network architectures on ADE20K object/component segmentation task. The rectangular contour indicates the best indices.

	Network	Backbone	Backbone depth	Decoder	Object segmentation accuracy (%)	Component segmentation accuracy (%)	Segmentation time (ms)
1	FCN [30]	ResNet	50	FCN	71.32	40.81	333
2	PSPNet [28]	ResNet	50	PPM	80.04	47.23	483
3	UPerNet [29]	ResNet	50	PPM + FPN	80.23	48.30	496
4	UPerNet [29]	ResNet	101	PPM + FPN	81.01	48.71	604