search

sicara / easy-few-shot-learning

Ready-to-use code and tutorial notebooks to boost your way into few-shot learning for image classification.
MIT License
961 stars 133 forks source link

Adding more backbones #62

Closed anish9 closed 1 year ago

anish9 commented 1 year ago

Hi @ebennequin, Thanks for this elegant code base, some questions(can be a feature request) 1) Can we add new backbones like (ViT, densenet, Convnext etc...)? 2) Building functionalities for model deployments?

ebennequin commented 1 year ago

Hi, thanks for the kind words.

  1. You can pass any backbone when building your FewShotClassifier, so you can use backbones like ViT, DenseNet, ConvNeXt from well-maintained libraries like timm. In this repo I added custom backbones that are hardly SOTA but are often used when benchmarking FSL.
  2. This repo is mainly targeted to research and education, not for production, so features for model deployment are not on the roadmap. However, I'd love to discuss it if you think we can add some useful stuff!
rlleshi commented 1 year ago

@anish9 I think model deployment is pretty straightforward. easyfsl is a very light library, so even if you don't want to convert models (e.g. to TorchServe or onxx) you can simply install the library & use SupportSetFolder for example.

Maybe adding a script to convert models would be a good idea though.