notesum.ai

Published at December 3

Class-wise Autoencoders Measure Classification Difficulty And Detect Label Mistakes

cs.LG

cs.CV

Released Date: December 3, 2024

Authors: Jacob Marks¹, Brent A. Griffin¹, Jason J. Corso²

Aff.: ¹Voxel51; ²University of Michigan

Arxiv: http://arxiv.org/pdf/2412.02596v1

Refer to caption

Dataset	$\overline{\chi}_{\infty}$	$\gamma_{1}$	$\gamma_{2}$	$R^{2}$
CIFAR-10	0.8411	19755.34	19875.82	0.9986
CIFAR-100	0.9478	23629.23	23660.84	0.9993
DeepWeeds	1.0071	3580.11	3578.83	0.9562
EuroSAT	0.8513	41763.83	41981.35	0.9984
Fashion-MNIST	0.8471	29558.87	29738.47	0.9983
MNIST	0.8566	25904.07	26032.33	0.9986
Stanford Dogs	0.9801	10564.68	10565.00	0.9986
FGVC-Aircraft	1.0027	290.81	291.17	0.9429
Mean	-	-	-	0.9864