TensorFlow Lite reference models

This page lists a number of reference neural networks generated using Edge Impulse. These can be used by silicon vendors to test the integration of common models that our users use with their neural network compilers or inferencing engines. Every model contains the TensorFlow SavedModel, a float32 tflite file, and two int8 tflite files - both with i8 and f32 inputs and outputs.
​Gestures model - a small fully connected network (33x40x20x10x4) built according to the Continuous motion recognition tutorial.
​Keywords model - a convolutional neural network with two conv1d layers built according to the Recognize sounds from audio tutorial.
​Visual model - a MobileNet v2 based model, built according to the Adding sight to your sensors tutorial.
License
All reference models are - like all code that is generated using Edge Impulse - licensed under the Apache 2.0 license.

Pre-built datasets - Previous

Keyword spotting

Next - Tips & Tricks

Increasing model performance

Last modified 18d ago

Export as PDF

Copy link

Contents

License