List your company
ResearchFormerly WizWand

Image Classification on CIFAR-100 60% symmetric noise (test)

73.58AccuracyFaMUS
34.153644.389354.62564.8607Apr 30, 2021May 6, 2021May 13, 2021May 19, 2021May 26, 2021Jun 1, 2021Jun 8, 2021
Updated 8mo ago

Evaluation results

MethodLinks
FaMUSBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0473.58——
DivideMixBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0472——
MentorMixBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0464.6——
MW-NetBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0461.71——
M-correctionBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0459.5——
MentorNetBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0458.8——
MixupBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0458.8——
L2RBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0450.81——
Negative Label SmoothingSmooth Rate=-62021.0646.46——
Negative Label SmoothingSmooth Rate=-42021.0646.38——
Negative Label SmoothingSmooth Rate=-22021.0644.88——
Negative Label SmoothingSmooth Rate=-0.82021.0644.1——
Negative Label SmoothingSmooth Rate=-12021.0643.85——
Negative Label SmoothingSmooth Rate=-0.42021.0643.7——
Label SmoothingSmooth Rate=0.42021.0641.57——
Label SmoothingSmooth Rate=0.62021.0640.49——
Label SmoothingSmooth Rate=0.82021.0639.5——
Vanilla LossSmooth Rate=02021.0638.11——
Co-teachingBackbone=PreAct ResNet-18, Noise Type=Symmetric, Noise Rate=60%2021.0435.67——
CEFine-tuning Protocol=FFT, Backbone=CLIP2024.09—81.0577.45
DEFTFine-tuning Protocol=Ours, Backbone=CLIP2024.09—85.8185.72
ELRFine-tuning Protocol=FFT, Backbone=CLIP2024.09—78.3478.34
GMMFine-tuning Protocol=FFT, Backbone=CLIP2024.09—85.2285.2
SCEFine-tuning Protocol=FFT, Backbone=CLIP2024.09—78.977.71