tflite_web 0.3.0

Run Tensorflow Lite models using TFLite JS and WebAssembly on browser

TFLite Web

Run TFLite models on Dart JS. It is packaged in a WebAssembly binary that runs in a browser

Getting Started

• Unpack tflite (link) into your web folder.
  Result structure:
  -- Web
  ├── tflite
  ──────├── tf-backend-cpu.js
  ──────├── tf-core.js
  ──────├── tf-tflite.min.js
  ──────├── tflite_web_...

• Initialize TFLite:

await TFLiteWeb.initialize();

This may take a couple of seconds

• Either Load Model from Url:

final model = await TFLiteModel.fromUrl(modelUrl);

• Or load from memory:

final model = await TFLiteModel.fromMemory(modelBytes);

• Create a tensor:

final input = Tensor(data, shape, dataType);

• Run the model:

final outputs = model.predict(input);

Depending on the model, the input can be a Tensor, a list of Tensors, or a Tensor map

Current Version:

• TF-JS: 4.11.0
• TF-JS TFLite: 0.0.1-alpha.10