TensorFlow Lite 2019 Roadmap

Updated: March 6th, 2019

The following represents a high level overview of our 2019 plan. You should be conscious that this roadmap may change at anytime relative to a range of factors and the order below does not reflect any type of priority. As a matter of principle, we typically prioritize issues that the majority of our users are asking for and so this list fundamentally reflects that.

We break our roadmap into four key segments: usability, performance, optimization and portability. We strongly encourage you to comment on our roadmap and provide us feedback in the TF Lite discussion groups and forums.

Usability

  • More ops coverage
    • Prioritize many more ops based on user feedback
  • Op versioning & signatures
    • Op kernels will get version numbers
    • Op kernels will be identifiable by signature
  • New Convertor
    • Implementing a new TensorFlow Lite convertor that will better handle graph conversion (i.e. control flow, conditionals etc) and replace TOCO
  • Continue to improve TF Select Ops
    • Support more types of conversion utilizing TF Selects such as hash tables, strings etc.
    • Support smaller binary size when using select TF ops via op stripping
  • LSTM / RNN support
    • Add full support of conversion for LSTMs and RNNs
  • Graph Visualization Tooling
    • Provide enhanced graph visualization tooling
  • Pre-and-post processing support
    • Add more support for pre-and-post processing of inference
  • Control Flow & Training on-device
    • Add support for control flow related ops
    • Add support for training on-device
  • New APIs
    • New C API as core for language bindings and most clients
    • Objective-C API for iOS
    • SWIFT API for iOS
    • Updated Java API for Android
    • C# Unity language bindings
  • Add more Models
    • Add more models to the support section of the site

Performance

  • More hardware delegates
    • Add support for more hardware delegates
  • Support NN API
    • Continually support and improve support for NN API
  • Framework Extensibility
    • Enable simplistic overwriting of CPU kernels with customized optimized versions
  • GPU Delegate
    • Continue to extend the total support ops for OpenGL and Metal ops
    • Open-source
  • Improve TFLite CPU performance
    • Optimizations for float and quantized models

Optimization

  • Model Optimization Toolkit
    • Post training quantization + hybrid kernels
    • Post Training quantization + fixed-point kernels
    • Training with quantization
  • More support for more techniques
    • RNN Support
    • Sparsity/Pruning
    • Lower bit-width support

Portability

  • Microcontroller Support
    • Add support for a range of 8-bit, 16-bit and 32-bit MCU architecture use cases for Speech and Image Classification