KRVPerera / FSL-MiniImageNet

Cross domain few-shot transfer learning from MiniImageNet to EuroSAT_RGB and CUB

CDFSL MiniImageNet to EuroSAT and CUB

Pretraining notebooks

• VGG
• ViT

Finetuning notebooks

EuroSAT_RGB

• ViT
• VGG

CUB

• ViT
• VGG

Accuracy of models

Pre-train on miniImageNet dataset

Model	Validation Accuracy
VGG19	82.84%
VisionTransformer	93.83%

Fine tuned on EuroSAT_RGB dataset

• Episodes : 20

Model	Average Test Accuracy
VGG19	78.94%
VisionTransformer	81.34%

Fine tuned on CUB dataset

• Episodes : 20

Model	Average Test Accuracy
VGG19	77.067%
VisionTransformer	81.34%

Fine tuned on CUB dataset

• Episodes : 50

Model	Average Test Accuracy
VGG19	73.41%
VisionTransformer	93.34%

Pretrained - models

RestNet152d - best_model_ModelResnet152dTimm_FixedTransforms.pth VGG19 - best_model_VGG19_fixedTransformers.pth VisionTransformer - best_model_VistionTransformerTimm_pc18_fixedTransformers.pth

You can download pretrained models from this location https://drive.google.com/drive/folders/1oD5kTRVJPSjhVz2C6cJeI1EXc2tmHMWq