sayakpaul / keras-convnext-conversion Goto Github PK

This repository holds the code that was used to populate the official ConvNeXt parameters [1, 2] into Keras ConvNeXt implementation. Most of the code is copied from here: https://github.com/sayakpaul/ConvNeXt-TF. Please refer to this repository for more comments, setup guides, etc.

The conversion was performed to aid this PR: keras-team/keras#16421.

Updates

• The above-mentioned PR's been merged and that means it's now available inside TensorFlow (nightly) for now. Here's the official documentation: https://www.tensorflow.org/api_docs/python/tf/keras/applications/convnext.
• Transfer learning code reference can be found here.

Execution

1. Install the Python dependencies: pip install -r requirements.txt.
2. Make sure you're at the root of the repository after cloning it.
3. Then create the required directories:

$ mkdir keras-applications
$ mkdir keras-applications/convnext

4. Then execute: python convert_all.py.

References

[1] ConvNeXt paper: https://arxiv.org/abs/2201.03545

[2] Official ConvNeXt code: https://github.com/facebookresearch/ConvNeXt