List your company
ResearchFormerly WizWand

Image Retrieval on SOP

78.56Precision@1MCL + PRISM
38.956849.238459.5269.8016Mar 30, 2021
Updated 8mo ago

Evaluation results

MethodLinks
MCL + PRISMNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0378.56
Memory Contrastive Loss (MCL)Noisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0375.61
MCL + PRISMNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0373.84
Soft TripleNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0373.63
Soft Triple + PRISMNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0370.99
CircleNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0370.33
Co-teaching w/ TemperatureNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0370.19
nSoftmaxNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0370
Soft Triple + PRISMNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0369.38
Memory Contrastive Loss (MCL)Noisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0368.71
Co-teaching w/ TemperatureNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0368.5
Contrastive LossNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0368.25
ProxyNCANoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0367.95
MSNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0367.32
Contrastive LossNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0364.27
Soft TripleNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0364.14
Co-teaching+Noisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0362.57
ProxyNCANoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0362.25
MSNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0362.17
Co-teachingNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0361.97
nSoftmaxNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0361.92
Co-teaching+Noisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0359.27
Co-teachingNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0358.08
FastAPNoisy Label Rate=25%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0352.18
FastAPNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0348.46
CircleNoisy Label Rate=50%, Label Noise Type=Small Cluster, Backbone=ResNet-50, Output Dimension=1282021.0340.48