Examples

We provide extensive examples to demonstrate Converter coverage and efficacy. Please check out converter_examples for detailed usages. We provide specific instructions for each example in the respective folder.

Evaluating supported model architectures

Tested_Models - This example evaluates common model architecture coverage by the individual conversion functions. It also summarizes known issues. We strongly recommend that users review this example first before running their conversion tasks.

Benchmarking quantized TensorFlow Lite models on an Android phone

Convert_Benchmark - This example demonstrates how to benchmark quantized TensorFlow Lite model performance (i.e. running time and memory usage) on an Android phone. Specifically, we converted models from a TensorFlow frozen protobuf file to a quantized .tflite file, and benchmarked their performance on an LG G6 mobile phone.

This example shows that models converted and quantized with Converter match the performance of the official quantized models provided in the TensorFlow repo.

Profiling performances of converted models using Auptimizer Profiler

Convert_Profiler - This example demonstrates how to use Auptimizer-Profiler to profile TensorFlow or ONNX model performance on CPU. Performance benchmarking scripts are provided for TensorFlow and ONNX.