Image Classification on ImageNet-LT (test)
82.3Top-1 Acc (All)Memory Attention Module
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Memory Attention ModuleRetrieval=true, Backbone=ViT-B16, Visual Encoder=Fine-tuned2023.04 | 82.3 | — | — | 85.4 | 81.5 | 76.4 | — | — | — | — | — | |
| RAC†Retrieval=true, Backbone=ViT-B16, Visual Encoder=Fine-tuned2023.04 | 80.5 | — | — | 85.9 | 79.3 | 69.3 | — | — | — | — | — | |
| GNM-PTBackbone=ViT-B/16, Modality=Visual-only2024.10 | 80.4 | — | — | 80.6 | 81.1 | 78.2 | — | — | — | — | — | |
| Memory Attention ModuleRetrieval=true, Backbone=ViT-B16, Visual Encoder=Frozen2023.04 | 78.3 | — | — | 80.6 | 77.5 | 74.5 | — | — | — | — | — | |
| LIFT w/ TTEBackbone=ViT-B/16, Learnable Params.=0.62M, #Epochs=10, Training Protocol=Fine-tuning foundation model2023.09 | 78.3 | — | — | 81.3 | 77.4 | 73.4 | — | — | — | — | — | |
| GMLBackbone=ViT-B/16, Learnable Params.=149.62M, #Epochs=100, Training Protocol=Fine-tuning with extra data2023.09 | 78 | — | — | — | — | — | — | — | — | — | — | |
| GMLBackbone=ViT-B/16, Modality=Supplementary with linguistic data2024.10 | 78 | — | — | — | — | — | — | — | — | — | — | |
| VL-LTRRetrieval=false, Backbone=ViT-B16, Visual Encoder=Fine-tuned2023.04 | 77.2 | — | — | 84.5 | 74.6 | 59.3 | — | — | — | — | — | |
| VL-LTRBackbone=ViT-B/16, Learnable Params.=149.62M, #Epochs=100, Training Protocol=Fine-tuning with extra data2023.09 | 77.2 | — | — | 84.5 | 74.6 | 59.3 | — | — | — | — | — | |
| VL-LTRBackbone=ViT-B/16, Modality=Supplementary with linguistic data2024.10 | 77.2 | — | — | 84.5 | 74.6 | 59.3 | — | — | — | — | — | |
| VPTBackbone=ViT-B/16, Modality=Visual-only2024.10 | 77.2 | — | — | 79.5 | 76.5 | 72.8 | — | — | — | — | — | |
| LIFTBackbone=ViT-B/16, Learnable Params.=0.62M, #Epochs=10, Training Protocol=Fine-tuning foundation model2023.09 | 77 | — | — | 80.2 | 76.1 | 71.5 | — | — | — | — | — | |
| RAC†Retrieval=true, Backbone=ViT-B16, Visual Encoder=Frozen2023.04 | 76.7 | — | — | 80.9 | 76 | 67.5 | — | — | — | — | — | |
| Mean k-NNRetrieval=true, Backbone=ViT-B16, Visual Encoder=Frozen2023.04 | 75.8 | — | — | 75.9 | 75.8 | 75.7 | — | — | — | — | — | |
| BALLADBackbone=ViT-B/16, Learnable Params.=149.62M, #Epochs=50+10, Training Protocol=Fine-tuning foundation model2023.09 | 75.7 | — | — | 79.1 | 74.5 | 69.8 | — | — | — | — | — | |
| BALLADBackbone=ViT-B/16, Modality=Visual-only2024.10 | 75.7 | — | — | 79.1 | 74.5 | 69.8 | — | — | — | — | — | |
| MLP ClassifierRetrieval=false, Backbone=ViT-B16, Visual Encoder=Frozen2023.04 | 75.2 | — | — | 80.1 | 74.1 | 66.9 | — | — | — | — | — | |
| Linear ClassifierRetrieval=false, Backbone=ViT-B16, Visual Encoder=Frozen2023.04 | 73.5 | — | — | 76.5 | 72.6 | 66.5 | — | — | — | — | — | |
| DecoderBackbone=ViT-B/16, Learnable Params.=21.26M, #Epochs=~18, Training Protocol=Fine-tuning foundation model2023.09 | 73.2 | — | — | — | — | — | — | — | — | — | — | |
| DecoderBackbone=ViT-B/16, Modality=Visual-only2024.10 | 73.2 | — | — | — | — | — | — | — | — | — | — | |
| BCL + ConCutMixBackbone=ResNeXt-101, Epochs=4002024.07 | 61.8 | — | — | 72.1 | 59.4 | 41.1 | — | — | — | — | — | |
| BCL + ConCutMixBackbone=ResNeXt-50, Epochs=4002024.07 | 61.3 | — | — | 72.1 | 58.4 | 40.8 | — | — | — | — | — | |
| LiVTBackbone=ViT-B/16, Learnable Params.=85.80M, #Epochs=100, Training Protocol=Training from scratch2023.09 | 60.9 | — | — | 73.6 | 56.4 | 41 | — | — | — | — | — | |
| GPaCoBackbone=ResNeXt-101, Epochs=4002024.07 | 60.8 | — | — | — | — | — | — | — | — | — | — | |
| PaCoRetrieval=false, Backbone=ResNext-101, Visual Encoder=Fine-tuned2023.04 | 60 | — | — | 68.2 | 58.7 | 41 | — | — | — | — | — | |
| PaCoBackbone=ResNeXt-101, Epochs=4002024.07 | 60 | — | — | 68.2 | 58.7 | 41 | — | — | — | — | — | |
| ResComBackbone=ResNeXt-502022.03 | 59.2 | — | — | 69.5 | 56.7 | 39.1 | — | — | — | — | — | |
| DeiT-LTBackbone=ViT-B, Teacher=PaCo+SAM2024.04 | 59.1 | — | — | 66.6 | 58.3 | 40 | — | — | — | — | — | |
| BCL + CutMixBackbone=ResNeXt-101, Epochs=4002024.07 | 59 | — | — | 67.7 | 57.3 | 40.9 | — | — | — | — | — | |
| BCL + ConCutMixBackbone=ResNet-50, Epochs=4002024.07 | 58.9 | — | — | 67.5 | 57.9 | 37.9 | — | — | — | — | — | |
| GPaCoBackbone=ResNeXt-50, Epochs=4002024.07 | 58.9 | — | — | — | — | — | — | — | — | — | — | |
| BCL + CutMixBackbone=ResNeXt-50, Epochs=4002024.07 | 58.9 | — | — | 68.4 | 57.5 | 37.1 | — | — | — | — | — | |
| SADEBackbone=ResNeXt-502021.07 | 58.8 | — | — | 66.5 | 57 | 43.5 | — | — | — | — | — | |
| BCL + MixupBackbone=ResNeXt-101, Epochs=4002024.07 | 58.7 | — | — | 67.4 | 56.9 | 40.8 | — | — | — | — | — | |
| BCLBackbone=ResNeXt-101, Epochs=4002024.07 | 58.6 | — | — | 67.4 | 56.9 | 40.2 | — | — | — | — | — | |
| GPaCoBackbone=ResNet-50, Epochs=4002024.07 | 58.5 | — | — | — | — | — | — | — | — | — | — | |
| BCL + MixupBackbone=ResNeXt-50, Epochs=4002024.07 | 58.5 | — | — | 67.8 | 57.2 | 37 | — | — | — | — | — | |
| BCLBackbone=ResNeXt-50, Epochs=4002024.07 | 58.3 | — | — | 67.8 | 56.8 | 36.9 | — | — | — | — | — | |
| PaCoBackbone=ResNeXt-502022.03 | 58.2 | — | — | 67.5 | 56.9 | 36.7 | — | — | — | — | — | |
| PaCoBackbone=ResNeXt-50, Epochs=4002024.07 | 58.2 | — | — | 67.5 | 56.9 | 36.7 | — | — | — | — | — | |
| BS + CMOBackbone=ResNet-50, Training Epochs=400, Data Augmentation=RandAugment [10]2021.12 | 58 | — | — | 67 | 55 | 44.2 | — | — | — | — | — | |
| BCL + CMOBackbone=ResNeXt-50, Epochs=4002024.07 | 57.9 | — | — | 65.2 | 56.4 | 42.9 | — | — | — | — | — | |
| BCL + CutMixBackbone=ResNet-50, Epochs=4002024.07 | 57.8 | — | — | 66.4 | 56.5 | 38.1 | — | — | — | — | — | |
| BCL + CMOBackbone=ResNeXt-101, Epochs=4002024.07 | 57.8 | — | — | 64.4 | 56.1 | 45.8 | — | — | — | — | — | |
| PaCo+SAMBackbone=ResNet502024.04 | 57.5 | — | — | 62.1 | 58.8 | 39.3 | — | — | — | — | — | |
| BCL + MixupBackbone=ResNet-50, Epochs=4002024.07 | 57.5 | — | — | 66.1 | 56.3 | 37.4 | — | — | — | — | — | |
| NCLBackbone=ResNet-50, Learnable Params.=23.51M, #Epochs=400, Training Protocol=Training from scratch2023.09 | 57.4 | — | — | — | — | — | — | — | — | — | — | |
| BCLBackbone=ResNet-50, Epochs=4002024.07 | 57.2 | — | — | 66.1 | 55.9 | 37.1 | — | — | — | — | — | |
| BCLBackbone=ResNet502024.04 | 57.1 | — | — | 67.9 | 54.2 | 36.6 | — | — | — | — | — | |
| PaCoBackbone=ResNet-50, Training Epochs=400, Data Augmentation=RandAugment [10]2021.12 | 57 | — | — | 65 | 55.7 | 38.2 | — | — | — | — | — | |
| PaCoBackbone=ResNet-50, Learnable Params.=23.51M, #Epochs=400, Training Protocol=Training from scratch2023.09 | 57 | — | — | — | — | — | — | — | — | — | — | |
| PaCoBackbone=ResNet-50, Epochs=4002024.07 | 57 | — | — | 65 | 55.7 | 38.2 | — | — | — | — | — | |
| RIDE(4 experts)Backbone=ResNeXt-502022.03 | 56.8 | — | — | 68.2 | 53.8 | 36 | — | — | — | — | — | |
| ACEBackbone=ResNeXt-502021.07 | 56.6 | — | — | 71.7 | 54.6 | 23.5 | — | — | — | — | — | |
| SMCBackbone=ResNet-50, training setup=presented in [34]2022.11 | 56.6 | — | — | 66.9 | 54.2 | 36.1 | — | — | — | — | — | |
| BCL + CMOBackbone=ResNet-50, Epochs=4002024.07 | 56.6 | — | — | 63.4 | 55.6 | 41.1 | — | — | — | — | — | |
| RIDEBackbone=ResNeXt-502021.07 | 56.3 | — | — | 68 | 52.9 | 35.1 | — | — | — | — | — | |
| ResLTBackbone=ResNeXt-50, One-stage=true, Data Augmentation=Strong (mixup and autoaugment), Knowledge distillation=false2021.01 | 56.1 | — | — | 63.6 | 55.7 | 38.9 | — | — | — | — | — | |
| RIDEbells_and_whistles=true2023.02 | 56.1 | — | — | 67.9 | 52.3 | 36 | — | — | — | — | — | |
| SSDBackbone=ResNeXt-502022.03 | 56 | — | — | 66.8 | 53.1 | 35.4 | — | — | — | — | — | |
| BCLBackbone=ResNet-50, training setup=presented in [34]2022.11 | 56 | — | — | — | — | — | — | — | — | — | — | |
| BCLBackbone=ResNet-50, Learnable Params.=23.51M, #Epochs=100, Training Protocol=Training from scratch2023.09 | 56 | — | — | — | — | — | — | — | — | — | — | |
| CBDENSBackbone=ResNet502024.04 | 55.6 | — | — | 68.5 | 52.7 | 29.2 | — | — | — | — | — | |
| DeiT-LTBackbone=ViT-B, Teacher=L-D-SAM2024.04 | 55.6 | — | — | 65.2 | 54 | 37.1 | — | — | — | — | — | |
| ImbSAMBackbone=ResNet502024.04 | 55.3 | — | — | 63.2 | 53.7 | 38.3 | — | — | — | — | — | |
| BSBackbone=ResNet-50, Training Epochs=400, Data Augmentation=RandAugment [10]2021.12 | 55 | — | — | 66.7 | 52.9 | 33 | — | — | — | — | — | |
| PC (Ours)bells_and_whistles=false2023.02 | 54.9 | — | — | 63.5 | 50.8 | 42.7 | — | — | — | — | — | |
| RIDE(3 exps)Backbone=ResNet502024.04 | 54.9 | — | — | 66.2 | 51.7 | 34.9 | — | — | — | — | — | |
| GCLBackbone=ResNet502024.04 | 54.5 | — | — | 63 | 52.7 | 37.1 | — | — | — | — | — | |
| PaCoBackbone=ResNeXt-502021.07 | 54.4 | — | — | 63.2 | 51.6 | 39.2 | — | — | — | — | — | |
| PaCObells_and_whistles=true2023.02 | 54.4 | — | — | 63.2 | 51.6 | 39.2 | — | — | — | — | — | |
| WD + WD + Maxbells_and_whistles=false2023.02 | 53.9 | — | — | 62.5 | 50.4 | 41.5 | — | — | — | — | — | |
| DRO-LTbells_and_whistles=false2023.02 | 53.5 | — | — | 64 | 49.8 | 33.1 | — | — | — | — | — | |
| DiVEBackbone=ResNeXt-502022.03 | 53.1 | — | — | 64.1 | 50.5 | 31.5 | — | — | — | — | — | |
| SAFABackbone=ResNet502024.04 | 53.1 | — | — | 63.8 | 49.9 | 33.4 | — | — | — | — | — | |
| L-D-SAMBackbone=ResNet502024.04 | 53.1 | — | — | 62 | 52.1 | 32.8 | — | — | — | — | — | |
| DiVE2021.03 | 53.1 | — | — | 64.06 | 50.41 | 31.46 | — | — | — | — | — | |
| LADEBackbone=ResNeXt-502022.03 | 53 | — | — | 65.1 | 48.9 | 33.4 | — | — | — | — | — | |
| ResLTBackbone=ResNeXt-50, One-stage=true, Knowledge distillation=false2021.01 | 52.9 | — | — | 63 | 50.5 | 35.5 | — | — | — | — | — | |
| DisAlignbells_and_whistles=false2023.02 | 52.9 | — | — | 61.3 | 52.2 | 31.4 | — | — | — | — | — | |
| DisalignBackbone=ResNet502024.04 | 52.9 | — | — | 61.3 | 52.2 | 31.4 | — | — | — | — | — | |
| DisAlignBackbone=ResNet-50, Learnable Params.=23.51M, #Epochs=90, Training Protocol=Training from scratch2023.09 | 52.9 | — | — | 61.3 | 52.2 | 31.4 | — | — | — | — | — | |
| SMCBackbone=ResNet-502022.11 | 52.7 | — | — | 61.8 | 49.8 | 37.9 | — | — | — | — | — | |
| MISLASBackbone=ResNet502024.04 | 52.7 | — | — | 62.9 | 50.7 | 34.3 | — | — | — | — | — | |
| MiSLASBackbone=ResNet-50, Learnable Params.=23.51M, #Epochs=180+10, Training Protocol=Training from scratch2023.09 | 52.7 | — | — | 62.9 | 50.7 | 34.3 | — | — | — | — | — | |
| TSCBackbone=ResNet-502022.11 | 52.4 | — | — | 63.5 | 49.7 | 30.4 | — | — | — | — | — | |
| TSCBackbone=ResNet502024.04 | 52.4 | — | — | 63.5 | 49.7 | 30.4 | — | — | — | — | — | |
| Balanced SoftmaxBackbone=ResNeXt-502021.07 | 52.3 | — | — | 64.1 | 48.2 | 33.4 | — | — | — | — | — | |
| LADEBackbone=ResNeXt-502021.07 | 52.3 | — | — | 64.4 | 47.7 | 34.3 | — | — | — | — | — | |
| CMOBackbone=ResNet-502022.11 | 52.3 | — | — | 62 | 49.1 | 36.7 | — | — | — | — | — | |
| Balanced SoftmaxBackbone=ResNeXt-502022.03 | 52.1 | — | — | 63.6 | 48.4 | 32.9 | — | — | — | — | — | |
| ELF(LDAM) + DRWBackbone=ResNet-50, Epochs=2002020.06 | 52 | — | — | 64.3 | 47.9 | 31.4 | -10 | — | — | — | — | |
| Causal NormBackbone=ResNeXt-502022.03 | 52 | — | — | 65.2 | 47.7 | 29.8 | — | — | — | — | — | |
| Causal NormBackbone=ResNeXt-50, One-stage=true, Knowledge distillation=false2021.01 | 51.9 | — | — | 64.9 | 48 | 28.9 | — | — | — | — | — | |
| TDEresults_source=original paper2021.03 | 51.8 | — | — | 62.7 | 48.8 | 31.6 | — | — | — | — | — | |
| τ-normalizedBackbone=ResNeXt-50, One-stage=true, Knowledge distillation=false2021.01 | 51.5 | — | — | 60.9 | 49.2 | 36.8 | — | — | — | — | — | |
| LWSBackbone=ResNeXt-50, One-stage=false, Knowledge distillation=false2021.01 | 51.5 | — | — | 63.3 | 48.4 | 32 | — | — | — | — | — | |
| KCLBackbone=ResNet-502022.11 | 51.5 | — | — | 61.8 | 49.4 | 30.9 | — | — | — | — | — | |
| MISLASBackbone=ResNeXt-502021.07 | 51.4 | — | — | 62 | 49.1 | 32.8 | — | — | — | — | — | |
| LABackbone=ResNet-50, Learnable Params.=23.51M, #Epochs=90, Training Protocol=Training from scratch2023.09 | 51.1 | — | — | — | — | — | — | — | — | — | — |