5-way Few-shot Image Classification on FC100 (test)
54.361-shot AccuracySemFew
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SemFewBackbone=ResNet-122023.11 | 54.36 | 62.79 | — | |
| SemFew-TransBackbone=Swin-T2023.11 | 54.27 | 65.02 | — | |
| MCT (Pair)Backbone=ResNet-12, Inference Protocol=Transductive2020.02 | 51.27 | 62.59 | — | |
| MCT (Instance)Backbone=ResNet-12, Inference Protocol=Transductive2020.02 | 51.16 | 63.28 | — | |
| Transductive fine-tuningBackbone=WRN-28-10, Pre-training split=train + val2019.09 | 50.44 | 65.74 | — | |
| SP-CLIPBackbone=Visformer-T, Params=10.0M, FLOPs=1.3 × 10^92023.03 | 48.53 | 61.55 | — | |
| SP-CLIPBackbone=Visformer-T2023.11 | 48.53 | 61.55 | — | |
| Transductive fine-tuningBackbone=ResNet-12, Pre-training split=train + val2019.09 | 48.29 | 63.38 | — | |
| FewTUREBackbone=Swin-T2023.11 | 47.68 | 63.81 | — | |
| MetaOpt SVMBackbone=ResNet-12*, Pre-training split=train + val2019.09 | 47.2 | 62.5 | — | |
| MetaOptNetBackbone=ResNet-122023.11 | 47.2 | 55.5 | — | |
| SP-SBERTBackbone=Visformer-T, Params=10.0M, FLOPs=1.3 × 10^92023.03 | 47.03 | 61.03 | — | |
| SP-GloVeBackbone=Visformer-T, Params=10.0M, FLOPs=1.3 × 10^92023.03 | 46.69 | 61.18 | — | |
| SKD-GEN1Backbone=ResNet-12, Training Set=train+val2020.06 | 46.5 | 63.1 | — | |
| AlignBackbone=WRN-28-10, Params=36.5M, FLOPs=3.7 × 10^102023.03 | 45.83 | 59.74 | — | |
| SKD-GEN0Backbone=ResNet-12, Training Set=train+val2020.06 | 45.3 | 62.2 | — | |
| Support-based initializationBackbone=WRN-28-10, Pre-training split=train + val2019.09 | 45.08 | 60.05 | — | |
| MCI (Instance)Backbone=ResNet-12, Inference Protocol=Inductive2020.02 | 44.69 | 60.33 | — | |
| RFS-distillBackbone=ResNet-12, Training Set=train+val2020.06 | 44.6 | 60.9 | — | |
| DistillBackbone=ResNet-12, Params=12.5M, FLOPs=3.5 × 10^92023.03 | 44.6 | 60.9 | — | |
| Fine-tuningBackbone=WRN-28-10, Pre-training split=train + val2019.09 | 43.8 | 64.4 | — | |
| infoPatchBackbone=ResNet-12, Params=12.5M, FLOPs=3.5 × 10^92023.03 | 43.8 | 58 | — | |
| Pre-trainBackbone=Visformer-T, Params=10.0M, FLOPs=1.3 × 10^92023.03 | 43.77 | 59.48 | — | |
| SSFormersBackbone=ResNet-122021.09 | 43.72 | 58.92 | — | |
| H-META-NASSize=55.52K2021.09 | 43.29 | 56.86 | — | |
| Fine-tuningBackbone=WRN-28-10, Inference Protocol=Transductive2020.02 | 43.16 | 57.57 | — | |
| Transductive fine-tuningBackbone=WRN-28-10, Pre-training split=train2019.09 | 43.16 | 57.57 | — | |
| Fine-tuningBackbone=WRN-28-102021.09 | 43.16 | 57.57 | — | |
| Support-based initializationBackbone=ResNet-12, Pre-training split=train + val2019.09 | 43.11 | 58.16 | — | |
| MCI (Pair)Backbone=ResNet-12, Inference Protocol=Inductive2020.02 | 43.01 | 59.67 | — | |
| Fine-tuningBackbone=ResNet-12, Pre-training split=train + val2019.09 | 42.84 | 63.1 | — | |
| RFS-simpleBackbone=ResNet-12, Training Set=train+val2020.06 | 42.6 | 59.1 | — | |
| RFSBackbone=ResNet-122023.11 | 42.6 | 59.1 | — | |
| MTLBackbone=ResNet-122021.06 | 42.4 | 57.7 | — | |
| MABASBackbone=ResNet-12, Params=12.5M, FLOPs=3.5 × 10^92023.03 | 42.31 | 57.56 | — | |
| MABASBackbone=ResNet-122021.09 | 42.31 | 57.56 | — | |
| MABASBackbone=ResNet-122023.11 | 42.31 | 58.16 | — | |
| RényiCLmulti-crop data augmentation=true, number of episodes=2000, harder augmentations=true2022.08 | 42.1 | 60.8 | — | |
| DenseBackbone=ResNet-12, Inference Protocol=Inductive2020.02 | 42.04 | 57.05 | — | |
| Transductive fine-tuningBackbone=ResNet-12, Pre-training split=train2019.09 | 41.89 | 54.96 | — | |
| ProtoNetBackbone=ResNet-122023.11 | 41.54 | 57.08 | — | |
| Meta-AdaMBackbone=ResNet-122023.11 | 41.12 | 56.14 | — | |
| MetaOpt-SVMBackbone=ResNet-12, Inference Protocol=Inductive2020.02 | 41.1 | 55.5 | — | |
| MetaOptNetBackbone=ResNet-12, Training Set=train+val2020.06 | 41.1 | 55.5 | — | |
| MetaOpt SVMBackbone=ResNet-12*, Pre-training split=train2019.09 | 41.1 | 55.5 | — | |
| MetaOptNetBackbone=ResNet-122021.06 | 41.1 | 55.5 | — | |
| MetaOptNetBackbone=ResNet-12, Params=12.5M, FLOPs=3.5 × 10^92023.03 | 41.1 | 55.5 | — | |
| MetaOptNetBackbone=ResNet-122021.09 | 41.1 | 55.5 | — | |
| CLSAmulti-crop data augmentation=true, number of episodes=2000, harder augmentations=true2022.08 | 41.09 | 58.38 | — | |
| T-NAS++Size=null2021.09 | 40.4 | 54.6 | — | |
| Transductive fine-tuningBackbone=conv (64)x4, Pre-training split=train + val2019.09 | 40.33 | 55.6 | — | |
| Support-based initializationBackbone=conv (64)x4, Pre-training split=train + val2019.09 | 40.25 | 54.53 | — | |
| TADAMBackbone=ResNet-12, Inference Protocol=Inductive2020.02 | 40.1 | 56.1 | — | |
| TADAMBackbone=ResNet-12, Training Set=train+val2020.06 | 40.1 | 56.1 | — | |
| TADAMBackbone=ResNet-12, Pre-training split=train2019.09 | 40.1 | 56.1 | — | |
| TADAMBackbone=ResNet-122021.06 | 40.1 | 56.1 | — | |
| TADAMBackbone=ResNet-122021.09 | 40.1 | 56.1 | — | |
| TADAMBackbone=ResNet-122023.11 | 40.1 | 56.1 | — | |
| T-NASSize=null2021.09 | 39.7 | 53.1 | — | |
| Fine-tuningBackbone=conv (64)x4, Pre-training split=train + val2019.09 | 39.09 | 54.83 | — | |
| MAML++Size=70.09K2021.09 | 38.7 | 52.9 | — | |
| Fine-tuningBackbone=WRN-28-10, Pre-training split=train2019.09 | 38.25 | 57.19 | — | |
| MAMLSize=70.09K2021.09 | 38.1 | 50.4 | — | |
| Fine-tuningBackbone=ResNet-12, Pre-training split=train2019.09 | 37.52 | 55.39 | — | |
| Prototypical Networks+Backbone=ResNet-12, Training Set=train+val2020.06 | 37.5 | 52.5 | — | |
| ProtoNetBackbone=ResNet-122021.06 | 37.5 | 52.5 | — | |
| ProtoNetBackbone=ResNet-12, Params=12.5M, FLOPs=3.5 × 10^92023.03 | 37.5 | 52.5 | — | |
| Prototypical NetBackbone=ResNet-122021.09 | 37.5 | 52.5 | — | |
| Support-based initializationBackbone=WRN-28-10, Pre-training split=train2019.09 | 36.82 | 49.72 | — | |
| Transductive fine-tuningBackbone=conv (64)x4, Pre-training split=train2019.09 | 36.62 | 50.24 | — | |
| RényiCLmulti-crop data augmentation=false, number of episodes=2000, harder augmentations=true2022.08 | 36.31 | 53.39 | — | |
| Support-based initializationBackbone=conv (64)x4, Pre-training split=train2019.09 | 36.07 | 48.72 | — | |
| Support-based initializationBackbone=ResNet-12, Pre-training split=train2019.09 | 35.51 | 48.26 | — | |
| Fine-tuningBackbone=conv (64)x4, Pre-training split=train2019.09 | 35.46 | 49.43 | — | |
| Prototypical Networks+Backbone=64-64-64-64, Training Set=train+val2020.06 | 35.3 | 48.6 | — | |
| Prototypical NetBackbone=Conv-64F2021.09 | 35.3 | 48.6 | — | |
| CLSAmulti-crop data augmentation=false, number of episodes=2000, harder augmentations=true2022.08 | 34.46 | 49.16 | — | |
| InfoMinmulti-crop data augmentation=false, number of episodes=2000, harder augmentations=true2022.08 | 31.8 | 45.09 | — | |
| Barlow TwinsBackbone=ResNet-12, Shots=1-shot, Paradigm=Unsupervised2022.10 | — | — | 37.9 | |
| Barlow TwinsBackbone=ResNet-12, Shots=5-shot, Paradigm=Unsupervised2022.10 | — | — | 54.1 | |
| Barlow TwinsNumber of shots=1-shot2024.02 | — | — | 37.9 | |
| Barlow TwinsNumber of shots=5-shot2024.02 | — | — | 54.1 | |
| Baseline++Backbone=ResNet-12, Shots=1-shot, Paradigm=Supervised2022.10 | — | — | 41.3 | |
| Baseline++Backbone=ResNet-12, Shots=5-shot, Paradigm=Supervised2022.10 | — | — | 58.7 | |
| BECLRNumber of shots=1-shot2024.02 | — | — | 45.21 | |
| BECLRNumber of shots=5-shot2024.02 | — | — | 60.02 | |
| BYOLBackbone=ResNet-12, Shots=1-shot, Paradigm=Unsupervised2022.10 | — | — | 37.2 | |
| BYOLBackbone=ResNet-12, Shots=5-shot, Paradigm=Unsupervised2022.10 | — | — | 52.8 | |
| CACTUS-MAMLBackbone=ResNet-12, Shots=1-shot, Paradigm=Unsupervised2022.10 | — | — | 31.3 | |
| CACTUS-MAMLBackbone=ResNet-12, Shots=5-shot, Paradigm=Unsupervised2022.10 | — | — | 45.7 | |
| Deep EigenmapsNumber of shots=1-shot2024.02 | — | — | 39.7 | |
| Deep EigenmapsNumber of shots=5-shot2024.02 | — | — | 57.9 | |
| Deep Laplacian eigenmapsBackbone=ResNet-12, Shots=1-shot, Paradigm=Unsupervised2022.10 | — | — | 39.7 | |
| Deep Laplacian eigenmapsBackbone=ResNet-12, Shots=5-shot, Paradigm=Unsupervised2022.10 | — | — | 57.9 | |
| HMSNumber of shots=1-shot2024.02 | — | — | 37.88 | |
| HMSNumber of shots=5-shot2024.02 | — | — | 53.68 | |
| LF2CSNumber of shots=1-shot2024.02 | — | — | 37.2 | |
| LF2CSNumber of shots=5-shot2024.02 | — | — | 52.8 | |
| MAMLBackbone=ResNet-12, Shots=1-shot, Paradigm=Supervised2022.10 | — | — | 36.9 | |
| MAMLBackbone=ResNet-12, Shots=5-shot, Paradigm=Supervised2022.10 | — | — | 51.2 |