5-way 1-shot classification on ImageNet mini
69.6Top-1 Accuracy (ACC_1)IbM2
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| IbM2Pretraining & Meta-training=PMF [22], Backbone=ViT-B/16, IbM2=true2023.05 | 69.6 | 95.6 | 5.1 | 77.6 | 87.1 | — | |
| FEAT-Res12+1S-DAug-3Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 69.25 | — | — | — | — | — | |
| FEAT-Res12+1S-DAug-2Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 69.04 | — | — | — | — | — | |
| DFRBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 67.74 | — | — | — | — | — | |
| PMFPretraining & Meta-training=PMF [22], Backbone=ResNet50, IbM2=false2023.05 | 67.2 | 94.9 | 5.3 | 75.2 | 86.3 | — | |
| FEAT-Res12+1S-DAug-1Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 67.08 | — | — | — | — | — | |
| PMFPretraining & Meta-training=PMF [22], Backbone=ViT-B/16, IbM2=false2023.05 | 66.9 | 94.9 | 5.4 | 75.5 | 86 | — | |
| FEAT-Res12Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 66.78 | — | — | — | — | — | |
| IbM2Pretraining & Meta-training=PMF [22], Backbone=ResNet50, IbM2=true2023.05 | 66.1 | 95.3 | 5.2 | 74.6 | 86.9 | — | |
| DeepEMDBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 65.91 | — | — | — | — | — | |
| ProtoNet-Res12+1S-DAug-3Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 64.94 | — | — | — | — | — | |
| ConstellationNetBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 64.89 | — | — | — | — | — | |
| ProtoNet-Res12+1S-DAug-2Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 64.61 | — | — | — | — | — | |
| FEAT-Res12 (re-impl.)Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 63.31 | — | — | — | — | — | |
| IbM2Pretraining & Meta-training=PMF [22], Backbone=ViT-S/16, IbM2=true2023.05 | 63.2 | 94.4 | 5.6 | 74.7 | 85.1 | — | |
| Meta-BaselineBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 63.17 | — | — | — | — | — | |
| SLA-AGBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 62.93 | — | — | — | — | — | |
| ProtoNet-Res12+1S-DAug-1Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 62.9 | — | — | — | — | — | |
| Meta-MaxUpBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 62.81 | — | — | — | — | — | |
| MetaOptNetBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 62.64 | — | — | — | — | — | |
| ProtoNet-Res12Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 62.39 | — | — | — | — | — | |
| PMFPretraining & Meta-training=PMF [22], Backbone=ViT-S/16, IbM2=false2023.05 | 61.3 | 93.4 | 6.1 | 72.4 | 83.4 | — | |
| ProtoNet + TRAMLBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 60.31 | — | — | — | — | — | |
| ProtoNet-Res12 (re-impl.)Backbone=Res12, Evaluation Protocol=Inductive2026.01 | 60.01 | — | — | — | — | — | |
| Classifier-BaselineBackbone=Res12, Evaluation Protocol=Inductive2026.01 | 58.91 | — | — | — | — | — | |
| OracleSelection Strategy=Oracle2025.12 | 46.1 | — | — | — | — | — | |
| IbM2Pretraining & Meta-training=S2M2R [28], Backbone=WRN-28-10, IbM2=true2023.05 | 32.5 | 65.8 | 10 | 40.5 | 51.1 | — | |
| IbM2Pretraining & Meta-training=Meta-Baseline [11], Backbone=ResNet12, IbM2=true2023.05 | 32.3 | 63 | 10.2 | 35.6 | 47.9 | — | |
| S2M2RPretraining & Meta-training=S2M2R [28], Backbone=WRN-28-10, IbM2=false2023.05 | 32 | 65.4 | 10 | 39.3 | 50.7 | — | |
| Source-ValSelection Strategy=Source Validation2025.12 | 29.4 | — | — | — | — | — | |
| Meta-BaselinePretraining & Meta-training=Meta-Baseline [11], Backbone=ResNet12, IbM2=false2023.05 | 26.7 | 62.9 | 10.4 | 34.3 | 47.4 | — | |
| De-DiffusionLLM=Llama2 70B, Task induction=false2023.11 | — | 64.8 | — | — | — | — | |
| De-DiffusionLLM=Llama2 70B, Task induction=true2023.11 | — | 87.9 | — | — | — | — | |
| De-DiffusionLLM=PaLM 2-S, Task induction=false2023.11 | — | 66.4 | — | — | — | — | |
| De-DiffusionLLM=PaLM 2-S, Task induction=true2023.11 | — | 88.6 | — | — | — | — | |
| De-DiffusionLLM=PaLM 2-L, Task induction=false2023.11 | — | 71.8 | — | — | — | — | |
| De-DiffusionLLM=PaLM 2-L, Task induction=true2023.11 | — | 97 | — | — | — | — | |
| FrozenLLM=Frozen, Task induction=false2023.11 | — | 0.9 | — | — | — | — | |
| FrozenLLM=Frozen, Task induction=true2023.11 | — | 33.8 | — | — | — | — | |
| LQAELLM=GPT3.5, Task induction=false2023.11 | — | 1 | — | — | — | — | |
| LQAELLM=GPT3.5, Task induction=true2023.11 | — | 45.9 | — | — | — | — | |
| P>M>FEvaluation protocol=Closed-form2023.11 | — | 95.3 | — | — | — | — | |
| SPAEPALMLLM=PaLM 2-L, Task induction=false2023.11 | — | 23.6 | — | — | — | — | |
| SPAEPALMLLM=PaLM 2-L, Task induction=true2023.11 | — | 67 | — | — | — | — |