Long-Tailed Image Classification on iNaturalist 2018
85.2AccuracyLIFT
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| LIFTBackbone=ViT-L/14, Learnable Params.=6.37M, #Epochs=20, Test-Time Ensembling (TTE)=true2023.09 | 85.2 | — | 80.2 | 85.1 | 86.6 | |
| LIFTBackbone=ViT-L/14, Learnable Params.=6.37M, #Epochs=202023.09 | 84.4 | — | 79.3 | 84.6 | 85.5 | |
| VL-LTRBackbone=ViT-Base, CLIP weights=true, Input size=384x3842021.11 | 81 | — | — | — | — | |
| DeiT-B/16Backbone=DeiT-B/16, Input size=384x3842021.11 | 79.5 | — | — | — | — | |
| BCL + ConCutMixBackbone=ResNet-152, Training Epochs=4002024.07 | 78.5 | — | 79.4 | 78.5 | 78.3 | |
| GPaCoBackbone=ResNet-152, Training Epochs=4002024.07 | 78.1 | — | — | — | — | |
| VL-LTRBackbone=ViT-Base, CLIP weights=true2021.11 | 76.8 | — | — | — | — | |
| BCLBackbone=ResNet-152, Training Epochs=4002024.07 | 76.8 | — | 78.4 | 77.2 | 76.3 | |
| BCL + ConCutMixBackbone=ResNet-50, Training Epochs=4002024.07 | 76 | — | 76.1 | 75.9 | 76.2 | |
| GPaCoBackbone=ResNet-50, Training Epochs=4002024.07 | 75.4 | — | — | — | — | |
| PaCoBackbone=ResNet-1522021.11 | 75.2 | — | — | — | — | |
| PaCoBackbone=ResNet-152, Training Epochs=4002024.07 | 75.2 | — | 75 | 75.5 | 74.7 | |
| DeiT-LTBackbone=ViT-B, Teacher=PaCo+SAM2024.04 | 75.1 | — | 70.3 | 75.2 | 76.2 | |
| VL-LTRBackbone=ResNet-50, CLIP weights=true2021.11 | 74.6 | — | — | — | — | |
| BCLBackbone=ResNet-50, Training Epochs=4002024.07 | 74 | — | 71.4 | 73.6 | 74.9 | |
| CBDENSBackbone=ResNet502024.04 | 73.6 | — | 75.9 | 74.7 | 71.5 | |
| PaCo+SAMBackbone=ResNet502024.04 | 73.4 | — | 66.3 | 73.6 | 75.2 | |
| PaCoBackbone=ResNet-502021.11 | 73.2 | — | — | — | — | |
| DeiT-B/16Backbone=DeiT-B/162021.11 | 73.2 | — | — | — | — | |
| PaCoBackbone=ResNet-50, Training Epochs=4002024.07 | 73.2 | — | 70.3 | 73.2 | 73.6 | |
| TADEBackbone=ResNet-502021.11 | 72.9 | — | — | — | — | |
| DeiT-LTBackbone=ViT-B, Teacher=L-D-SAM2024.04 | 72.9 | — | 69 | 73.3 | 73.3 | |
| RIDE (4 experts)Backbone=ResNet-502021.11 | 72.6 | — | — | — | — | |
| BaselineBackbone=ResNet-50, CLIP weights=true2021.11 | 72.6 | — | — | — | — | |
| τ-normalized classifierBackbone=ResNet-152, Epochs=2002019.10 | 72.5 | — | — | — | — | |
| ResLTBackbone=ResNet-502021.11 | 72.3 | — | — | — | — | |
| DecoderBackbone=ViT-L/14, Learnable Params.=39.79M, #Epochs=~52023.09 | 72.3 | — | 65.5 | 73.2 | 73 | |
| RIDE(3 exps)Backbone=ResNet502024.04 | 72.2 | — | 70.2 | 72.2 | 72.7 | |
| Learnable Weight Scaling (LWS)Backbone=ResNet-152, Epochs=2002019.10 | 72.1 | — | — | — | — | |
| MiSLAS (under-conf.)Backbone=ResNet-50, confidence_mode=under-conf.2021.04 | 71.6 | 7.67 | — | — | — | |
| MISLASBackbone=ResNet502024.04 | 71.6 | — | 73.2 | 72.4 | 70.4 | |
| SSDBackbone=ResNet-502021.11 | 71.5 | — | — | — | — | |
| cRTBackbone=ResNet-152, Epochs=2002019.10 | 71.2 | — | — | — | — | |
| τ-normalizedBackbone=ResNet-50, CLIP weights=true2021.11 | 71.2 | — | — | — | — | |
| ImbSAMBackbone=ResNet502024.04 | 71.1 | — | 68.2 | 72.5 | 72.9 | |
| LWSBackbone=ResNet-50, CLIP weights=true2021.11 | 71 | — | — | — | — | |
| GCLBackbone=ResNet502024.04 | 71 | — | 67.5 | 71.3 | 71.5 | |
| LWS+mixup (under-conf.)Backbone=ResNet-50, confidence_mode=under-conf.2021.04 | 70.9 | 9.41 | — | — | — | |
| DisalignBackbone=ResNet502024.04 | 70.6 | — | 69 | 71.1 | 70.2 | |
| Remix+DRWBackbone=ResNet-502021.04 | 70.5 | — | — | — | — | |
| cRT+mixupBackbone=ResNet-502021.04 | 70.2 | 1.79 | — | — | — | |
| L-D-SAMBackbone=ResNet502024.04 | 70.1 | — | 64.1 | 70.5 | 71.2 | |
| CRTBackbone=ResNet-50, CLIP weights=true2021.11 | 69.9 | — | — | — | — | |
| TSCBackbone=ResNet502024.04 | 69.7 | — | 72.6 | 70.6 | 67.8 | |
| BBN (include mixup)Backbone=ResNet-502021.04 | 69.6 | — | — | — | — | |
| BBNBackbone=ResNet-502021.11 | 69.6 | — | — | — | — | |
| Learnable Weight Scaling (LWS)Backbone=ResNet-50, Epochs=2002019.10 | 69.5 | — | — | — | — | |
| LWSBackbone=ResNet-502021.11 | 69.5 | — | — | — | — | |
| τ-normalized classifierBackbone=ResNet-50, Epochs=2002019.10 | 69.3 | — | — | — | — | |
| τ-normalizedBackbone=ResNet-502021.11 | 69.3 | — | — | — | — | |
| Learnable Weight Scaling (LWS)Backbone=ResNet-152, Epochs=902019.10 | 69.1 | — | — | — | — | |
| smDRAGONBackbone=ResNet-502021.11 | 69.1 | — | — | — | — | |
| JointBackbone=ResNet-152, Epochs=2002019.10 | 69 | — | — | — | — | |
| τ-normalized classifierBackbone=ResNet-152, Epochs=902019.10 | 68.8 | — | — | — | — | |
| cRTBackbone=ResNet-152, Epochs=902019.10 | 68.5 | — | — | — | — | |
| LDAM+DRWBackbone=ResNet-502019.10 | 68 | — | — | — | — | |
| LDAM+DRWBackbone=ResNet-502021.04 | 68 | — | — | — | — | |
| LDAM+DRWBackbone=ResNet-502021.11 | 68 | — | — | — | — | |
| cRTBackbone=ResNet-50, Epochs=2002019.10 | 67.6 | — | — | — | — | |
| CRTBackbone=ResNet-502021.11 | 67.6 | — | — | — | — | |
| Nearest Class Mean (NCM)Backbone=ResNet-152, Epochs=2002019.10 | 67.3 | — | — | — | — | |
| Learnable Weight Scaling (LWS)Backbone=ResNet-50, Epochs=902019.10 | 65.9 | — | — | — | — | |
| JointBackbone=ResNet-50, Epochs=2002019.10 | 65.8 | — | — | — | — | |
| τ-normalized classifierBackbone=ResNet-50, Epochs=902019.10 | 65.6 | — | — | — | — | |
| t-NormBackbone=ResNet502024.04 | 65.6 | — | 65.6 | 65.3 | 65.9 | |
| NCMBackbone=ResNet-50, CLIP weights=true2021.11 | 65.3 | — | — | — | — | |
| cRTBackbone=ResNet-50, Epochs=902019.10 | 65.2 | — | — | — | — | |
| c-RTBackbone=ResNet502024.04 | 65.2 | — | 69 | 66 | 63.2 | |
| JointBackbone=ResNet-152, Epochs=902019.10 | 65 | — | — | — | — | |
| LDAMBackbone=ResNet-502019.10 | 64.6 | — | — | — | — | |
| Nearest Class Mean (NCM)Backbone=ResNet-50, Epochs=2002019.10 | 63.1 | — | — | — | — | |
| NCMBackbone=ResNet-502021.11 | 63.1 | — | — | — | — | |
| Nearest Class Mean (NCM)Backbone=ResNet-152, Epochs=902019.10 | 61.9 | — | — | — | — | |
| JointBackbone=ResNet-50, Epochs=902019.10 | 61.7 | — | — | — | — | |
| CB-FocalBackbone=ResNet-502019.10 | 61.1 | — | — | — | — | |
| CB-FocalBackbone=ResNet-502021.04 | 61.1 | — | — | — | — | |
| CB-FocalBackbone=ResNet-502021.11 | 61.1 | — | — | — | — | |
| DeiT-IIIBackbone=ViT-B2024.04 | 61 | — | 72.9 | 62.8 | 55.8 | |
| Nearest Class Mean (NCM)Backbone=ResNet-50, Epochs=902019.10 | 58.2 | — | — | — | — | |
| ViTBackbone=ViT-B2024.04 | 54.2 | — | 64.3 | 53.9 | 52.1 | |
| CLIPBackbone=ViT-L/14, Mode=Zero-shot2023.09 | 5.8 | — | 11.1 | 5.4 | 4.9 | |
| Zero-Shot CLIPBackbone=ResNet-50, CLIP weights=true, Evaluation Protocol=Zero-Shot2021.11 | 3.4 | — | — | — | — |