Long-tailed recognition on Places-LT (test)
49.5Accuracy (Overall)BALLAD
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BALLADBackbone=ViT-B/16, #Pretrain=N2021.11 | 49.5 | 49.3 | 50.2 | 48.4 | — | |
| BALLADBackbone=ResNet-50×16, #Pretrain=N2021.11 | 49.3 | 49.4 | 50.5 | 46.6 | — | |
| BALLADBackbone=ResNet-101, #Pretrain=N2021.11 | 47.9 | 48 | 48.6 | 46 | — | |
| GLAFine-tuning protocol=Prompt Tuning (PT)2023.10 | 47.2 | 47 | 47.1 | 47.8 | — | |
| BALLADBackbone=ResNet-50, #Pretrain=N2021.11 | 46.5 | 46.7 | 48 | 42.7 | — | |
| GLAFine-tuning protocol=End-to-End (E2E)2023.10 | 46.1 | 47.3 | 45.4 | 45.3 | — | |
| BALLADFine-tuning protocol=End-to-End (E2E)2023.10 | 45.7 | 48.7 | 44.7 | 42.2 | — | |
| WiSE-FTFine-tuning protocol=End-to-End (E2E)2023.10 | 44.7 | 46 | 44 | 43.9 | — | |
| BSFine-tuning protocol=End-to-End (E2E)2023.10 | 44.3 | 46.7 | 44.4 | 39.5 | — | |
| BSFine-tuning protocol=Prompt Tuning (PT)2023.10 | 42.8 | 43.4 | 42.5 | 42.3 | — | |
| MDCSBackbone=ResNet-152, Pre-trained=ImageNet, RandAug=true2023.08 | 42.4 | 43.1 | 42.9 | 36.3 | — | |
| NCLBackbone=ResNet-152, Pre-trained=ImageNet, RandAug=true2023.08 | 41.8 | — | — | — | — | |
| PaCo + BatchFormerBackbone=ResNet-1522022.03 | 41.6 | 44 | 43.1 | 33.7 | — | |
| LAFine-tuning protocol=Prompt Tuning (PT)2023.10 | 41.5 | 43.9 | 40.6 | 39.4 | — | |
| PaCoBackbone=ResNet-152, #Pretrain=Y2021.11 | 41.2 | 37.5 | 47.2 | 33.9 | — | |
| PaCoBackbone=ResNet-152, #Pretrain=Y, RandAugment=true2021.11 | 41.2 | 36.1 | 47.9 | 35.3 | — | |
| PaCoBackbone=ResNet-1522022.03 | 41.2 | 37.5 | 47.2 | 33.9 | — | |
| PaCoBackbone=ResNet-152, Pre-trained=ImageNet, RandAug=true2023.08 | 41.2 | 36.1 | 47.9 | 35.3 | — | |
| GLAFine-tuning protocol=Linear Probe (LP)2023.10 | 41.1 | 43.6 | 40.2 | 38.8 | — | |
| MiSLASBackbone=ResNet-152, Pre-trained=ImageNet, RandAug=false2023.08 | 40.4 | 39.6 | 43.3 | 36.1 | — | |
| BSCEBackbone=ResNet-152, Pre-trained=ImageNet, RandAug=true2023.08 | 40.2 | — | — | — | — | |
| ResLTBackbone=ResNet-152, #Pretrain=Y2021.11 | 39.8 | 39.8 | 43.6 | 31.4 | — | |
| ResLTBackbone=ResNet-152, Pre-trained=ImageNet, RandAug=false2023.08 | 39.8 | 39.8 | 43.6 | 31.4 | — | |
| LDAM-DRS-RSGBackbone=ResNet-1522021.06 | 39.3 | 41.9 | 41.4 | 32 | — | |
| LAFine-tuning protocol=End-to-End (E2E)2023.10 | 39.3 | 42.2 | 39.3 | 34 | — | |
| Ensemble of class-balanced experts (All)Backbone=ResNet-1522020.04 | 38.9 | 43.6 | 39.9 | 27.7 | — | |
| BBNBackbone=ResNet-1522021.06 | 38.7 | 42.5 | 40.3 | 30.6 | — | |
| LDAM-DRSBackbone=ResNet-1522021.06 | 38.6 | 43.3 | 38.3 | 30.7 | — | |
| Balanced SoftmaxBackbone=ResNet-152, #Pretrain=Y2021.11 | 38.6 | 42 | 39.3 | 30.5 | — | |
| LWSFine-tuning protocol=End-to-End (E2E)2023.10 | 38.6 | 41.2 | 39 | 32.8 | — | |
| BALMS + BatchFormerBackbone=ResNet-152, trained_with_released_code=true2022.03 | 38.2 | 39.5 | 38.3 | 35.7 | — | |
| τ-normalized classifierBackbone=ResNet-152, Pre-trained=ImageNet2019.10 | 37.9 | 37.8 | 40.7 | 31.8 | — | |
| tau-normalizedBackbone=ResNet-1522021.06 | 37.9 | 37.8 | 40.7 | 31.8 | — | |
| τ-normalizedBackbone=ResNet-152, #Pretrain=Y2021.11 | 37.9 | 37.8 | 40.7 | 31.8 | — | |
| T-normalizedBackbone=ResNet-1522022.03 | 37.9 | 37.8 | 40.7 | 31.8 | — | |
| τ-normBackbone=ResNet-152, Pre-trained=ImageNet, RandAug=false2023.08 | 37.9 | 37.8 | 40.7 | 31.8 | — | |
| BALMSBackbone=ResNet-152, trained_with_released_code=true2022.03 | 37.8 | 41.4 | 38.8 | 29.1 | — | |
| Learnable Weight Scaling (LWS)Backbone=ResNet-152, Pre-trained=ImageNet2019.10 | 37.6 | 40.6 | 39.1 | 28.6 | — | |
| LWSBackbone=ResNet-152, #Pretrain=Y2021.11 | 37.6 | 40.6 | 39.1 | 28.6 | — | |
| cRTBackbone=ResNet-152, Pre-trained=ImageNet2019.10 | 36.7 | 42 | 37.6 | 24.9 | — | |
| CRTBackbone=ResNet-152, #Pretrain=Y2021.11 | 36.7 | 42 | 37.6 | 24.9 | — | |
| ERMFine-tuning protocol=Prompt Tuning (PT)2023.10 | 36.5 | 47.7 | 32.6 | 24.9 | — | |
| Nearest Class Mean (NCM)Backbone=ResNet-152, Pre-trained=ImageNet2019.10 | 36.4 | 40.4 | 37.1 | 27.3 | — | |
| OLTRBackbone=ResNet-152, Pre-trained=ImageNet2019.10 | 35.9 | 44.7 | 37 | 25.3 | — | |
| Liu et al.Backbone=ResNet-152, source=taken directly from [24]2020.04 | 35.9 | 44.7 | 37 | 25.3 | — | |
| OLTRBackbone=ResNet-1522021.06 | 35.9 | 44.7 | 37 | 25.3 | — | |
| OLTRBackbone=ResNet-152, #Pretrain=Y2021.11 | 35.9 | 44.7 | 37 | 25.3 | — | |
| OLTRBackbone=ResNet-1522022.03 | 35.9 | 44.7 | 37 | 25.3 | — | |
| OLTRBackbone=ResNet-152, Pre-trained=ImageNet, RandAug=false2023.08 | 35.9 | 44.7 | 37 | 25.3 | — | |
| UniformBackbone=ResNet-1522020.04 | 35.6 | 41.3 | 35.5 | 25.2 | — | |
| Lifted LossBackbone=ResNet-152, Pre-trained=ImageNet2019.10 | 35.2 | 41.1 | 35.4 | 24 | — | |
| Lifted LossBackbone=ResNet-1522020.04 | 35.2 | 41.1 | 35.4 | 24 | — | |
| Liu et al.Backbone=ResNet-152, source=obtained from author's code2020.04 | 35.2 | 41.4 | 37.1 | 19.2 | — | |
| Lifted LossBackbone=ResNet-1522021.06 | 35.2 | 41.1 | 35.4 | 24 | — | |
| Range LossBackbone=ResNet-152, Pre-trained=ImageNet2019.10 | 35.1 | 41.1 | 35.4 | 23.2 | — | |
| Range LossBackbone=ResNet-1522020.04 | 35.1 | 41.1 | 35.4 | 23.2 | — | |
| Range LossBackbone=ResNet-1522021.06 | 35.1 | 41.1 | 35.4 | 23.2 | — | |
| FSLWBackbone=ResNet-152, Pre-trained=ImageNet2019.10 | 34.9 | 43.9 | 29.9 | 29.5 | — | |
| FSLwFBackbone=ResNet-152, source=taken directly from [24]2020.04 | 34.9 | 43.9 | 29.9 | 29.5 | — | |
| FSLWFBackbone=ResNet-1522021.06 | 34.9 | 43.9 | 29.9 | 29.5 | — | |
| LAFine-tuning protocol=Linear Probe (LP)2023.10 | 34.9 | 39.2 | 33.7 | 29.5 | — | |
| BSFine-tuning protocol=Linear Probe (LP)2023.10 | 34.9 | 39.6 | 34 | 28.3 | — | |
| Focal LossBackbone=ResNet-152, Pre-trained=ImageNet2019.10 | 34.6 | 41.1 | 34.8 | 22.4 | — | |
| Focal LossBackbone=ResNet-1522020.04 | 34.6 | 41.1 | 34.8 | 22.4 | — | |
| GANBackbone=ResNet-1522020.04 | 34.6 | 42.7 | 33.3 | 22.5 | — | |
| Focal LossBackbone=ResNet-1522021.06 | 34.6 | 41.1 | 34.8 | 22.4 | — | |
| Ensemble of class-balanced experts (Experts)Backbone=ResNet-1522020.04 | 34.2 | 37.2 | 35.3 | 26.3 | — | |
| ERMFine-tuning protocol=End-to-End (E2E)2023.10 | 31.3 | 46.3 | 27.4 | 12.5 | — | |
| JointBackbone=ResNet-152, Pre-trained=ImageNet2019.10 | 30.2 | 45.7 | 27.3 | 8.2 | — | |
| BaselineBackbone=ResNet-1522020.04 | 29.5 | 45.4 | 25.6 | 9 | — | |
| ERMFine-tuning protocol=Linear Probe (LP)2023.10 | 26.9 | 43.9 | 21.1 | 9 | — | |
| BALMSBackbone=ResNet-152, Pre-training=ImageNet2021.12 | — | — | — | — | 38.7 | |
| ERMBackbone=ResNet-152, Pre-training=ImageNet2021.12 | — | — | — | — | 29.9 | |
| LADEBackbone=ResNet-152, Pre-training=ImageNet2021.12 | — | — | — | — | 38.8 | |
| MiSLASBackbone=ResNet-152, Pre-training=ImageNet2021.12 | — | — | — | — | 40.4 | |
| OPeNBackbone=ResNet-152, Pre-training=ImageNet2021.12 | — | — | — | — | 40.5 | |
| OversamplingBackbone=ResNet-152, Pre-training=ImageNet2021.12 | — | — | — | — | 38.1 |