TFLite Model Optimization and Benchmarking Application

This application is designed to convert Keras models to TensorFlow Lite (TFLite) models using quantization and to benchmark multiple models.

App Architecture

Directory Structure

• Benchmarks/: Contains all the code used for benchmarking models.

• Quantization_tools/: Contains the code to perform quantization on models.

• Models/: Contains the models you want to convert or quantize. Each model directory must follow the implementation examples (e.g., Mnist or CloudNet).

  • Each model directory should include a run.py script with a class representing your model that extends BaseModel. This class must have convert and benchmark functions to handle conversion to TFLite and benchmarking, respectively.

Main Script

• run_model.py: The main script of the application used to launch the processes.
  • Usage: python run_model.py {model_name} {action}
    • {model_name}: The name of the model directory (e.g., mnist).
    • {action}: The action to perform (benchmark or convert).

Usage Example

To start the benchmark function for the MNIST model, run:

python run_model.py mnist benchmark

Getting Started

1. Clone the repository:

   git clone https://github.com/Alex-Delaveau/TFLite-Optimize-Bench.git

2. Install the required dependencies:

   pip install -r requirements.txt

3. (Optional) If you want to work on the CloudNet model, install its specific dependencies:

   pip install -r models/cloudnet/requirements.txt

4. Run the desired model and action:

   python run_model.py {model_name} {action}

Replace {model_name} with the name of your model (e.g., mnist) and {action} with the desired action (benchmark or convert).