MobileViTv2

This repository contains the implementation of MobileViTv2 model in Keras 3.

Project Structure

• mobilevit_v2.py: Contains the main model implementation.
• configs.py: Contains the model configurations.
• test.py: Script for evaluating the model.
• utils/base_layers.py: Contains base layers.
• utils/linear_attention.py: Contains Separable Self-attention implementation.
• utils/mobilevit_v2_block.py: Contains MobileViTv2 block implementation.
• utils/utils.py: Contains utility functions.

Usage

For evaluating the model on an example images run the test.py script:

python test.py

This will:

• Sets the Keras 3 backend.
• Downloads an example image from the Web.
• Makes three different MobileViTv2 models with width multipliers α = 0.5, 1, 2 and downloads the pre-trained weights.
• Tests these three models on the example image.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgments

• The MobileViTv2 model architecture is inspired by the original MobileViTv2 paper.
• The code is heavily borrowed from this github repo.