notesum.ai

Published at November 8

Revisiting Network Perturbation for Semi-Supervised Semantic Segmentation

cs.CV

cs.AI

Released Date: November 8, 2024

Authors: Sien Li¹, Tao Wang², Ruizhe Hu³, Wenxi Liu¹

Aff.: ¹College of Computer and Data Science, Fuzhou University, Fuzhou 350108, China; ²School of Computer and Big Data, Minjiang University, Fuzhou 350108, China; ³School of Computer Science and Mathematics, Fujian University of Technology, Fuzhou 350118, China

Arxiv: http://arxiv.org/abs/2411.05307v1

Pascal VOC	Resolution	ResNet-101
Pascal VOC	Resolution	92	183	366	732
SupOnly	321×321	45.1	55.3	64.8	69.7
CPS [CVPR’21][4]	512×512	64.1	67.4	71.4	75.9
ST++ [CVPR’22] [32]	321×321	65.2	71.0	74.6	77.3
U²PL [CVPR’22] [25]	512×512	68.0	69.2	73.7	76.2
PS-MT [CVPR’22][18]	512×512	65.8	69.6	76.6	78.4
PCR [NeurIPS’22][30]	512×512	70.1	74.7	77.2	78.5
ESL [ICCV’23][20]	513×513	70.1	74.1	78.1	79.5
SemiCVT [ICCV’23][11]	513×513	68.6	71.2	75.0	78.5
Hunting Sparse [ICCV’23][24]	513×513	70.5	77.1	78.7	79.2
LogicDiag [ICCV’23][17]	513×513	73.3	76.7	77.9	79.4
3-cps [ICCV’23][16]	321×321	75.4	76.8	79.6	80.4
3-cps [ICCV’23][16]	513×513	75.7	77.7	80.1	80.9
iMas [CVPR’23][34]	513×513	70.0	75.3	79.1	80.2
augseg [CVPR’23][35]	513×513	71.1	75.5	78.8	80.3
UniMatch [CVPR’23][31]	321×321	75.2	77.2	78.8	79.9
MLPMatch	321×321	76.8	78.2	80.3	80.4
Gain	-	↑31.7	↑22.9	↑15.5	↑10.7