List your company

Long-Tailed Image Classification on iNaturalist 2018

85.2AccuracyLIFT
0.12822.21444.366.386Oct 21, 2019Aug 2, 2020May 16, 2021Feb 27, 2022Dec 10, 2022Sep 23, 2023Jul 6, 2024

Evaluation results

MethodLinks
LIFTBackbone=ViT-L/14, Learnable Params.=6.37M, #Epochs=20, Test-Time Ensembling (TTE)=true2023.0985.2—80.285.186.6
LIFTBackbone=ViT-L/14, Learnable Params.=6.37M, #Epochs=202023.0984.4—79.384.685.5
VL-LTRBackbone=ViT-Base, CLIP weights=true, Input size=384x3842021.1181————
DeiT-B/16Backbone=DeiT-B/16, Input size=384x3842021.1179.5————
BCL + ConCutMixBackbone=ResNet-152, Training Epochs=4002024.0778.5—79.478.578.3
GPaCoBackbone=ResNet-152, Training Epochs=4002024.0778.1————
VL-LTRBackbone=ViT-Base, CLIP weights=true2021.1176.8————
BCLBackbone=ResNet-152, Training Epochs=4002024.0776.8—78.477.276.3
BCL + ConCutMixBackbone=ResNet-50, Training Epochs=4002024.0776—76.175.976.2
GPaCoBackbone=ResNet-50, Training Epochs=4002024.0775.4————
PaCoBackbone=ResNet-1522021.1175.2————
PaCoBackbone=ResNet-152, Training Epochs=4002024.0775.2—7575.574.7
DeiT-LTBackbone=ViT-B, Teacher=PaCo+SAM2024.0475.1—70.375.276.2
VL-LTRBackbone=ResNet-50, CLIP weights=true2021.1174.6————
BCLBackbone=ResNet-50, Training Epochs=4002024.0774—71.473.674.9
CBDENSBackbone=ResNet502024.0473.6—75.974.771.5
PaCo+SAMBackbone=ResNet502024.0473.4—66.373.675.2
PaCoBackbone=ResNet-502021.1173.2————
DeiT-B/16Backbone=DeiT-B/162021.1173.2————
PaCoBackbone=ResNet-50, Training Epochs=4002024.0773.2—70.373.273.6
TADEBackbone=ResNet-502021.1172.9————
DeiT-LTBackbone=ViT-B, Teacher=L-D-SAM2024.0472.9—6973.373.3
RIDE (4 experts)Backbone=ResNet-502021.1172.6————
BaselineBackbone=ResNet-50, CLIP weights=true2021.1172.6————
τ-normalized classifierBackbone=ResNet-152, Epochs=2002019.1072.5————
ResLTBackbone=ResNet-502021.1172.3————
DecoderBackbone=ViT-L/14, Learnable Params.=39.79M, #Epochs=~52023.0972.3—65.573.273
RIDE(3 exps)Backbone=ResNet502024.0472.2—70.272.272.7
Learnable Weight Scaling (LWS)Backbone=ResNet-152, Epochs=2002019.1072.1————
MiSLAS (under-conf.)Backbone=ResNet-50, confidence_mode=under-conf.2021.0471.67.67———
MISLASBackbone=ResNet502024.0471.6—73.272.470.4
SSDBackbone=ResNet-502021.1171.5————
cRTBackbone=ResNet-152, Epochs=2002019.1071.2————
τ-normalizedBackbone=ResNet-50, CLIP weights=true2021.1171.2————
ImbSAMBackbone=ResNet502024.0471.1—68.272.572.9
LWSBackbone=ResNet-50, CLIP weights=true2021.1171————
GCLBackbone=ResNet502024.0471—67.571.371.5
LWS+mixup (under-conf.)Backbone=ResNet-50, confidence_mode=under-conf.2021.0470.99.41———
DisalignBackbone=ResNet502024.0470.6—6971.170.2
Remix+DRWBackbone=ResNet-502021.0470.5————
cRT+mixupBackbone=ResNet-502021.0470.21.79———
L-D-SAMBackbone=ResNet502024.0470.1—64.170.571.2
CRTBackbone=ResNet-50, CLIP weights=true2021.1169.9————
TSCBackbone=ResNet502024.0469.7—72.670.667.8
BBN (include mixup)Backbone=ResNet-502021.0469.6————
BBNBackbone=ResNet-502021.1169.6————
Learnable Weight Scaling (LWS)Backbone=ResNet-50, Epochs=2002019.1069.5————
LWSBackbone=ResNet-502021.1169.5————
τ-normalized classifierBackbone=ResNet-50, Epochs=2002019.1069.3————
τ-normalizedBackbone=ResNet-502021.1169.3————
Learnable Weight Scaling (LWS)Backbone=ResNet-152, Epochs=902019.1069.1————
smDRAGONBackbone=ResNet-502021.1169.1————
JointBackbone=ResNet-152, Epochs=2002019.1069————
τ-normalized classifierBackbone=ResNet-152, Epochs=902019.1068.8————
cRTBackbone=ResNet-152, Epochs=902019.1068.5————
LDAM+DRWBackbone=ResNet-502019.1068————
LDAM+DRWBackbone=ResNet-502021.0468————
LDAM+DRWBackbone=ResNet-502021.1168————
cRTBackbone=ResNet-50, Epochs=2002019.1067.6————
CRTBackbone=ResNet-502021.1167.6————
Nearest Class Mean (NCM)Backbone=ResNet-152, Epochs=2002019.1067.3————
Learnable Weight Scaling (LWS)Backbone=ResNet-50, Epochs=902019.1065.9————
JointBackbone=ResNet-50, Epochs=2002019.1065.8————
τ-normalized classifierBackbone=ResNet-50, Epochs=902019.1065.6————
t-NormBackbone=ResNet502024.0465.6—65.665.365.9
NCMBackbone=ResNet-50, CLIP weights=true2021.1165.3————
cRTBackbone=ResNet-50, Epochs=902019.1065.2————
c-RTBackbone=ResNet502024.0465.2—696663.2
JointBackbone=ResNet-152, Epochs=902019.1065————
LDAMBackbone=ResNet-502019.1064.6————
Nearest Class Mean (NCM)Backbone=ResNet-50, Epochs=2002019.1063.1————
NCMBackbone=ResNet-502021.1163.1————
Nearest Class Mean (NCM)Backbone=ResNet-152, Epochs=902019.1061.9————
JointBackbone=ResNet-50, Epochs=902019.1061.7————
CB-FocalBackbone=ResNet-502019.1061.1————
CB-FocalBackbone=ResNet-502021.0461.1————
CB-FocalBackbone=ResNet-502021.1161.1————
DeiT-IIIBackbone=ViT-B2024.0461—72.962.855.8
Nearest Class Mean (NCM)Backbone=ResNet-50, Epochs=902019.1058.2————
ViTBackbone=ViT-B2024.0454.2—64.353.952.1
CLIPBackbone=ViT-L/14, Mode=Zero-shot2023.095.8—11.15.44.9
Zero-Shot CLIPBackbone=ResNet-50, CLIP weights=true, Evaluation Protocol=Zero-Shot2021.113.4————