TFLite Flutter Helper Library brings TFLite Support Library and TFLite Support Task Library to Flutter and helps users to develop ML and deploy TFLite models onto mobile devices quickly without compromising on performance.
Follow the initial setup instructions given here
TFLite Helper depends on flutter image package internally for Image Processing.
The TensorFlow Lite Support Library has a suite of basic image manipulation methods such as crop
and resize. To use it, create an ImageProcessor
and add the required operations.
To convert the image into the tensor format required by the TensorFlow Lite interpreter,
create a TensorImage
to be used as input:
// Initialization code
// Create an ImageProcessor with all ops required. For more ops, please
// refer to the ImageProcessor Ops section in this README.
ImageProcessor imageProcessor = ImageProcessorBuilder()
.add(ResizeOp(224, 224, ResizeMethod.NEAREST_NEIGHBOUR))
.build();
// Create a TensorImage object from a File
TensorImage tensorImage = TensorImage.fromFile(imageFile);
// Preprocess the image.
// The image for imageFile will be resized to (224, 224)
tensorImage = imageProcessor.process(tensorImage);
Sample app: Image Classification
The TensorFlow Lite Support Library also defines a TensorAudio class wrapping some basic audio data processing methods.
TensorAudio tensorAudio = TensorAudio.create(
TensorAudioFormat.create(1, sampleRate), size);
tensorAudio.loadShortBytes(audioBytes);
TensorBuffer inputBuffer = tensorAudio.tensorBuffer;
Sample app: Audio Classification
// Create a container for the result and specify that this is a quantized model.
// Hence, the 'DataType' is defined as UINT8 (8-bit unsigned integer)
TensorBuffer probabilityBuffer =
TensorBuffer.createFixedSize(<int>[1, 1001], TfLiteType.uint8);
import 'package:tflite_flutter/tflite_flutter.dart';
try {
// Create interpreter from asset.
Interpreter interpreter =
await Interpreter.fromAsset("mobilenet_v1_1.0_224_quant.tflite");
interpreter.run(tensorImage.buffer, probabilityBuffer.buffer);
} catch (e) {
print('Error loading model: ' e.toString());
}
Developers can access the output directly through probabilityBuffer.getDoubleList()
.
If the model produces a quantized output, remember to convert the result.
For the MobileNet quantized model, the developer needs to divide each output value by 255
to obtain the probability ranging from 0 (least likely) to 1 (most likely) for each category.
Developers can also optionally map the results to labels. First, copy the text file containing labels into the module’s assets directory. Next, load the label file using the following code:
List<String> labels = await FileUtil.loadLabels("assets/labels.txt");
The following snippet demonstrates how to associate the probabilities with category labels:
TensorLabel tensorLabel = TensorLabel.fromList(
labels, probabilityProcessor.process(probabilityBuffer));
Map<String, double> doubleMap = tensorLabel.getMapWithFloatValue();
The design of the ImageProcessor allowed the image manipulation operations to be defined up front and optimised during the build process. The ImageProcessor currently supports three basic preprocessing operations:
int cropSize = min(_inputImage.height, _inputImage.width);
ImageProcessor imageProcessor = ImageProcessorBuilder()
// Center crop the image to the largest square possible
.add(ResizeWithCropOrPadOp(cropSize, cropSize))
// Resize using Bilinear or Nearest neighbour
.add(ResizeOp(224, 224, ResizeMethod.NEAREST_NEIGHBOUR))
// Rotation clockwise in 90 degree increments
.add(Rot90Op(rotationDegrees ~/ 90))
.add(NormalizeOp(127.5, 127.5))
.add(QuantizeOp(128.0, 1 / 128.0))
.build();
See more details here about normalization and quantization.
The TensorProcessor can be used to quantize input tensors or dequantize output tensors. For example, when processing a quantized output TensorBuffer, the developer can use DequantizeOp to dequantize the result to a floating point probability between 0 and 1:
// Post-processor which dequantize the result
TensorProcessor probabilityProcessor =
TensorProcessorBuilder().add(DequantizeOp(0, 1 / 255.0)).build();
TensorBuffer dequantizedBuffer =
probabilityProcessor.process(probabilityBuffer);
// Quantization Params of input tensor at index 0
QuantizationParams inputParams = interpreter.getInputTensor(0).params;
// Quantization Params of output tensor at index 0
QuantizationParams outputParams = interpreter.getOutputTensor(0).params;
Currently, Text based models like NLClassifier
, BertNLClassifier
and BertQuestionAnswerer
are available to use with the Flutter Task Library.
The Task Library's NLClassifier
API classifies input text into different categories, and is a versatile and configurable API that can handle most text classification models. Detailed guide is available here.
final classifier = await NLClassifier.createFromAsset('assets/$_modelFileName',
options: NLClassifierOptions());
List<Category> predictions = classifier.classify(rawText);
Sample app: Text Classification.
The Task Library BertNLClassifier
API is very similar to the NLClassifier
that classifies input text into different categories, except that this API is specially tailored for Bert related models that require Wordpiece and Sentencepiece tokenizations outside the TFLite model. Detailed guide is available here.
final classifier = await BertNLClassifier.createFromAsset('assets/$_modelFileName',
options: BertNLClassifierOptions());
List<Category> predictions = classifier.classify(rawText);
The Task Library BertQuestionAnswerer
API loads a Bert model and answers questions based on the content of a given passage. For more information, see the documentation for the Question-Answer model here. Detailed guide is available here.
final bertQuestionAnswerer = await BertQuestionAnswerer.createFromAsset('assets/$_modelFileName');
List<QaAnswer> answeres = bertQuestionAnswerer.answer(context, question);
Sample app: Bert Question Answerer Sample