notesum.ai

Published at November 8

GCI-ViTAL: Gradual Confidence Improvement with Vision Transformers for Active Learning on Label Noise

cs.CV

cs.AI

cs.LG

Released Date: November 8, 2024

Authors: Moseli Mots'oehli¹, kyungim Baek¹

Aff.: ¹Department of Information and Computer Sciences, University of Hawai'i at Manoa, Honolulu, HI 96822

Arxiv: http://arxiv.org/abs/2411.05939v1

Refer to caption

Noise Rate	Model	Active Learning Strategy
Noise Rate	Model	Random	Entropy	Margin	Hybrid	Model Delta	GCI-ViTAL
0	VGG19	0.8278	0.8288	0.8273	0.8370	0.8287	-
	ResNet34	0.8132	0.8319	0.8398	0.8348	0.8328	-
	ViTb16	0.9330	0.9320	0.9322	0.9338	0.9345	0.9338
	SwinV2b16	0.8946	0.8922	0.8945	0.8936	0.8936	0.8939
0.2	VGG19	0.7821	0.7564	0.7575	0.7433	0.7137	-
	ResNet34	0.7843	0.7496	0.7426	0.7499	0.7278	-
	ViTb16	0.9176	0.9174	0.9173	0.9117	0.9185	0.9286
	SwinV2b16	0.8828	0.8887	0.8837	0.8808	0.8779	0.8830
0.4	VGG19	0.7406	0.7280	0.6646	0.6651	0.6296	-
	ResNet34	0.7485	0.6691	0.6968	0.6496	0.6009	-
	ViTb16	0.9093	0.9020	0.9047	0.9114	0.9127	0.9152
	SwinV2b16	0.8757	0.8789	0.8841	0.8791	0.8829	0.8874
0.6	VGG19	0.7063	0.6450	0.6737	0.6612	0.4912	-
	ResNet34	0.7041	0.6234	0.6592	0.5575	0.4477	-
	ViTb16	0.9011	0.8677	0.8922	0.8933	0.8815	0.9117
	SwinV2b16	0.8777	0.8643	0.8738	0.8697	0.8670	0.8850