notesum.ai

Published at December 9

Active Learning with Context Sampling and One-vs-Rest Entropy for Semantic Segmentation

cs.CV

cs.LG

Released Date: December 9, 2024

Authors: Fei Wu¹, Pablo Marquez-Neila¹, Hedyeh Rafi-Tarii, Raphael Sznitman¹

Aff.: ¹University of Bern

Arxiv: http://arxiv.org/pdf/2412.06470v1

Refer to caption

Method		MONARCH	PASCAL VOC	ENDOVIS	CITYSCAPES	Average
		mean - max	mean - max	mean - max	mean - max	mean - max
random	RN50	67.6	75.1	74.0	62.0	69.7
BvSB [18]	RN50	67.2 - 71.9	74.7 - 77.0	74.9 - 75.5	65.4 - 65.1	70.6 - 72.4
revisiting SP [8]	RN50	68.6 - 73.7	75.9 - 76.5	75.2 - 74.8	66.2 - 66.7	71.5 - 72.9
Pixel Bal [17]	RN50	67.8 - 64.3	74.5 - 77.0	75.2 - 76.0	66.1 - 65.5	70.9 - 70.7
CBAL [7]	RN50	64.3 - 64.0	73.1 - 73.1	75.2 - 74.8	66.9 - 67.1	69.9 - 69.8
OREAL (ours)	RN50	66.2 - 69.7	73.6 - 77.2	74.8 - 76.2	67.9 - 68.0	70.6 - 72.8
random	RN101	74.5	75.4	72.2	59.7	70.5
BvSB [18]	RN101	76.2 - 77.2	74.0 - 77.3	73.1 - 74.5	62.8 - 63.3	71.5 - 73.1
revisiting SP [8]	RN101	76.9 - 75.5	75.3 - 77.0	72.9 - 72.9	63.7 - 65.5	72.2 - 72.7
Pixel Bal [17]	RN101	75.7 - 75.2	73.6 - 76.8	73.1 - 75.0	63.4 - 65.5	71.5 - 73.1
CBAL [7]	RN101	73.4 - 71.5	73.6 - 72.3	74.3 - 74.0	65.5 - 65.9	71.7 - 70.9
OREAL (ours)	RN101	74.9 - 74.7	74.9 - 77.9	73.8 - 74.8	66.3 - 66.3	72.5 - 73.4
random	ViT	57.2	76.2	75.8	65.7	68.7
BvSB [18]	ViT	66.4 - 72.1	72.4 - 76.8	76.2 - 78.0	69.0 - 68.7	71.0 - 73.9
revisiting SP [8]	ViT	68.0 - 72.0	72.1 - 76.3	76.6 - 76.9	69.5 - 69.5	71.6 - 73.7
Pixel Bal [17]	ViT	67.3 - 72.4	71.8 - 76.1	76.1 - 77.9	69.6 - 69.7	71.2 - 73.8
CBAL [7]	ViT	72.5 - 70.9	71.8 - 72.7	76.2 - 76.4	71.1 - 70.1	72.9 - 72.5
OREAL (ours)	ViT	74.2 - 72.4	73.2 - 76.5	76.3 - 76.9	68.5 - 69.6	73.1 - 73.9
Average	(Excl. random)	70.6 - 71.8	73.6 - 76.0	74.9 - 75.6	66.8 - 67.1	71.5 - 72.7