TFLite Web
Run TFLite models on Dart. 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);
- If you need to access the model from the assets in Flutter, you can either:
final url = AssetManager().getAssetUrl(asset);
final model = await TFLiteModel.fromUrl(url);
- Or:
final byteData = await rootBundle.load(asset);
final bytes = byteData.buffer.asUint8List();
model = await TFLiteModel.fromMemory(bytes);
- 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
Flutter
If you're wondering how to use this package with tflite_flutter, please refer to this article (link)
Current Version:
- TF-JS: 4.11.0
- TF-JS TFLite: 0.0.1-alpha.10