Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference

Benchmark Model Rank Results
few-shot-image-classification-on-cifar-fs-5P>M>F (P=DINO-ViT-base, M=ProtoNet)#9Accuracy: 84.3
few-shot-image-classification-on-cifar-fs-5-1P>M>F (P=DINO-ViT-base, M=ProtoNet)#4Accuracy: 92.2
few-shot-image-classification-on-meta-datasetP>M>F (P=DINO-ViT-base, M=ProtoNet)#2Accuracy: 84.75
few-shot-image-classification-on-mini-2P>M>F (P=DINO-ViT-base, M=ProtoNet)#3Accuracy: 95.3
few-shot-image-classification-on-mini-3P>M>F (P=DINO-ViT-base, M=ProtoNet)#3Accuracy: 98.4