search

Sha-Lab / FEAT

The code repository for "Few-Shot Learning via Embedding Adaptation with Set-to-Set Functions"
MIT License
418 stars 84 forks source link

Performance of instance embeddings without training #16

Closed mctigger closed 5 years ago

mctigger commented 5 years ago

Hi, can you tell me what the performance of the pretrained backbones is if their embeddings are not further adapted or trained in a few-shot setting, but only evaluated?

Best regards, Tim

Han-Jia commented 5 years ago

Hi, Tim,

For ConvNet, the pretrained embedding gets 1-shot performance near 46% (based on distance measure). For ResNet, it gets about 55%.

You can directly use eval_matchnet/protonet with the pretrained embedding to get the precise few-shot classification performance for the pretrained embedding.

mctigger commented 5 years ago

Thanks a lot!