By ranking predictions across progressively masked views, REM reduces model collapse in entropy-based test-time adaptation and improves CTTA accuracy on ImageNet-C, CIFAR10-C, and CIFAR100-C.
ImageNet-C Source Tent SAR ViDA REM ECE (%) ↓ 5.3 12.6 10.3 14.6 8.7 Error (%) ↓ 55.8 51.0 45.2 43.4 39.2 H
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Ranked Entropy Minimization for Continual Test-Time Adaptation
By ranking predictions across progressively masked views, REM reduces model collapse in entropy-based test-time adaptation and improves CTTA accuracy on ImageNet-C, CIFAR10-C, and CIFAR100-C.