Porting all models from everywhere to Keras to leverage multi-backend support.

Cause why not?🤷🏻‍♂️

Table of Contents

• Table of Contents
  • Progress
  • Updates
  • Quick Setup
    • Stable PyPi Package
    • OR
    • Latest Git Updates
  • Models Supported

Progress

• Working on adding MobileNet V4

Updates

1. [2025-03-12] Finished adding FastViT image classification model by Apple added. The model weights are available at url.
2. [2025-02-22] Working (v1) FastViT classification code added.
3. [2024-06-24] Released MobileViT v2 - All Image Classification variants.
4. [2024-05-15] Fixed MobileViT v1 - Now works will all 3 backends. 🎉🎉
5. [2024-05-04] Converted MobileViT to Keras 3 and released weights of all 3 variants.
   1. Jax backend currently not working, I'm working on a fix.
   2. Release: https://github.com/veb-101/keras-vision/releases/tag/v0.4

Quick Setup

Stable PyPi Package

pip install -U keras-vision

OR

Latest Git Updates

pip install git+https://github.com/veb-101/keras-vision.git

Models Supported

# No.	Models	Paper	Additional Materials	Example notebook	Weights URL
1	MobileViT-V1	MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer	Blogpost: Building MobileViT In Keras 3	Colab link	🤗
2	MobileViT-V2	Separable Self-attention for Mobile Vision Transformers	--	Colab link	🤗
2	FastViT	FastViT: A Fast Hybrid Vision Transformer using Structural Reparameterization	--	Colab link	🤗

Citation

• MobileViT-v1

@inproceedings{mehta2022mobilevit,
     title={MobileViT: Light-weight, General-purpose, and Mobile-friendly Vision Transformer},
     author={Sachin Mehta and Mohammad Rastegari},
     booktitle={International Conference on Learning Representations},
     year={2022}
}

• MobileViT-v2

Mehta, S., & Rastegari, M. (2022). Separable Self-attention for Mobile Vision Transformers. ArXiv. https://arxiv.org/abs/2206.02680

• FastViT

@inproceedings{vasufastvit2023,
  author = {Pavan Kumar Anasosalu Vasu and James Gabriel and Jeff Zhu and Oncel Tuzel and Anurag Ranjan},
  title = {FastViT:  A Fast Hybrid Vision Transformer using Structural Reparameterization},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  year = {2023}
}