How to contribute to this repository?

Feel free to contribute classifiers, backbones, functions and any enhancements.

Add a method/feature or fix a bug

We recommend using the following guidelines:

  1. fork the main branch of the latest LibFewShot;

  2. checkout a new branch，whose name should reflects the content intuitively, like add-method-ProtoNet of fix-doc-contribution;

  3. add a new method/feature or fix a bug;

  4. check and commit;

  5. create a pull request.

Note that if you add a new method, you need:

  1. test if the method works properly;

  2. provide a config file of this new method, and the corresponding 5-way 1-shot and 5-way 5-shot accuracy on the miniImageNet dataset.

Also, it will be better if you can provide:

  1. the 5-way 1-shot and 5-way 5-shot accuracy on other datasets (like tieredImageNet);

  2. model_best.pth of each setting on each dataset.

We will thank you for your contributions in README or other prominent places.

Use pre-commit to check code

Before committing the code, you may need to make sure that your code could pass black and flake test. We use pre-commit to do test and automatic code revision:

  1. first, install pre-commit;

cd <path-to-LibFewShot>
pip install pre-commit

  1. run pre-commit install;

  2. run pre-commit run --all-files;

  3. modify the code by the warning gived by pre-commit.

Pull request style

The title of your PR should like followings:

[Method] XXXX XXXX
# OR
[Feature] XXXX XXXX
# OR
[FIX] XXXX XXXX

The body of your PR should describe the main content of this PR in EN OR CN.