Tensor/IO is a lightweight, cross-platform library for on-device machine learning, bringing the power of TensorFlow and TensorFlow Lite to iOS, Android, and React Native applications. We now also support PyTorch on Android and will be bringing support for PyTorch to iOS soon. Tensor/IO does not implement any machine learning itself but works with an underlying library such as TensorFlow to simplify the process of deploying and using models on mobile phones.
Tensor/IO is above all a declarative interface to your model. Describe the input and output layers to your model using a plain-text language and Tensor/IO takes care of the transformations needed to prepare inputs for the model and to read outputs back out of it, allowing you to focus on what you know instead of a low-level C++ interface.
Tensor/IO runs on iOS and Android mobile phones, with bridging for React Native, and it runs the same underlying model on every OS without needing to convert models to CoreML or MLKit. You'll choose a specific backend for your use case, such as TensorFlow or TensorFlow Lite, and the library takes care of interacting with it in the language of your choice: Objective-C, Swift, Java, Kotlin, or JavaScript.
Prediction with Tensor/IO can often be done with as little as five lines of code. The TensorFlow Lite backend supports deep neural networks and a range of convolutional models, and the full TensorFlow backend supports almost any network you can build in python. Performance is impressive. MobileNet models execute inference on the iPhone X in ~30ms and can be run in real-time.
With support for the full TensorFlow backend you can train models on device and then export their updated weights, which can be immediately used in a prediction model. All without ever leaving the phone. Use the same declarative interface to specify your training inputs and outputs, evaluation metric, and training operation, and to inject placeholder values into your model for on-device hyperparameter tuning.
Given a TensorFlow Lite MobileNet ImageNet classification model that has been packaged into a Tensor/IO bundle (bundled here), the model's description looks like:
{
"name": "MobileNet V2 1.0 224",
"details": "MobileNet V2 with a width multiplier of 1.0 and an input resolution of 224x224. \n\nMobileNets are based on a streamlined architecture that have depth-wise separable convolutions to build light weight deep neural networks. Trained on ImageNet with categories such as trees, animals, food, vehicles, person etc. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications.",
"id": "mobilenet-v2-100-224-unquantized",
"version": "1",
"author": "Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam",
"license": "Apache License. Version 2.0 http://www.apache.org/licenses/LICENSE-2.0",
"model": {
"file": "mobilenet_v2_1.4_224.tflite",
"backend": "tflite",
"quantized": false,
"type": "image.classification.imagenet"
},
"inputs": [
{
"name": "image",
"type": "image",
"shape": [224,224,3],
"format": "RGB",
"normalize": {
"standard": "[-1,1]"
}
}
],
"outputs": [
{
"name": "classification",
"type": "array",
"shape": [1,1000],
"labels": "labels.txt"
}
]
}
Use the model on iOS:
UIImage *image = [UIImage imageNamed:@"example-image"];
TIOPixelBuffer *buffer = [[TIOPixelBuffer alloc] initWithPixelBuffer:image.pixelBuffer orientation:kCGImagePropertyOrientationUp];
TIOTFLiteModel *model = [TIOTFLiteModel modelWithBundleAtPath:path];
NSDictionary *inference = (NSDictionary*)[model runOn:buffer];
NSDictionary *classification = [inference[@"classification"] topN:5 threshold:0.1];
Or with Swift:
let image = UIImage(named: "example-image")!
let pixels = image.pixelBuffer()!
let value = pixels.takeUnretainedValue() as CVPixelBuffer
let buffer = TIOPixelBuffer(pixelBuffer:value, orientation: .up)
let model = TIOTFLiteModel.withBundleAtPath(path)!
let inference = model.run(on: buffer)
let classification = ((inference as! NSDictionary)["classification"] as! NSDictionary).topN(5, threshold: 0.1)
In Java on Android:
InputStream bitmap = getAssets().open("picture2.jpg");
Bitmap bMap = BitmapFactory.decodeStream(bitmap);
TIOModelBundleManager manager = new TIOModelBundleManager(getApplicationContext(), path);
TIOModelBundle bundle = manager.bundleWithId(bundleId);
TIOModel model = bundle.newModel();
model.load();
float[] result = (float[]) model.runOn(bMap);
String[] labels = ((TIOVectorLayerDescription)model.descriptionOfOutputAtIndex(0)).getLabels();
Or Kotlin:
val bitmap = assets.open("picture2.jpg")
val bMap = BitmapFactory.decodeStream(bitmap)
val manager = TIOModelBundleManager(applicationContext, path)
val bundle = manager.bundleWithId(bundleId)
val model = bundle.newModel()
model.load()
val result = model.runOn(bMap) as FloatArray
And in React Native:
import TensorioTflite from 'react-native-tensorio-tflite';
const { imageKeyFormat, imageKeyData, imageTypeAsset } = TensorioTflite.getConstants();
const imageAsset = Platform.OS === 'ios' ? 'elephant' : 'elephant.jpg';
TensorioTflite.load('image-classification.tiobundle', 'classifier');
TensorioTflite
.run('classifier', {
'image': {
[imageKeyFormat]: imageTypeAsset,
[imageKeyData]: imageAsset
}
})
.then(output => {
// @ts-ignore
return TensorioTflite.topN(5, 0.1, output['classification'])
})
.then(setResults)
.catch(error => {
console.log(error)
});
TensorioTflite.unload('classifier');
All Tensor/IO, Net Runner, and related code is open source under an Apache 2 license. Copyright doc.ai, 2018-present.
Part of this repository, a collection of jupyter notebooks showing how to build models that can be exported for inference and training on device with Tensor/IO. Also includes iOS and Android example applications showing how to use each of those models on device.
A full end-to-end walkthrough of preparing a model for on-device training in a Google Colab notebook and then running it on iOS and Android.
Our Objective-C++ implementation of Tensor/IO, with support for Swift. Requires iOS 12.0+. Older versions of the framework support iOS 9.3+ and have been tested on devices as old as a 5th generation iPod touch (2012). Support TF Lite and TensorFlow backends.
Net Runner is our iOS application environment for running and evaluating computer vision machine learning models packaged for Tensor/IO. Models may be run on live camera input or bulk evaluated against album photos. New models may be downloaded directly into the application. Net Runner is available for download in the iOS App Store.
Our Java implementation of Tensor/IO for Android. Requires Java 8. Kotlin compatible. Supports TF Lite, TensorFlow, and PyTorch backends.
Net Runner is our Android application environment for running computer vision models on device, including support for GPU acceleration.
Tensor/IO TF Lite for React Native
Our React Native bindings for Tensor/IO with a TF Lite backend. Use TF Lite models on both iOS and Android via a single React Native app.
Tensor/IO TensorFlow for React Native
Our React Native bindings for Tensor/IO with a full TensorFlow backend. This is a work in progress as we resolve some dependency conflicts with the build.
An expanding suite of python tools for working with Tensor/IO bundles, the format used to package models for running on device with Tensor/IO.
Our TensorFlow fork with fixes and additional ops enabled to support both training and inference on iOS and Android. See specifically the r2.0.doc.ai branch and our build script for composing the ios framework and our comprehensive readme for instructions on building tensorflow for Android.
Our on-device build of TensorFlow 2.0 supports models built in TF v1.13 - v2.2 and may work with some models built in later versions of TensorFlow.
Unofficial build of the full tensorflow.framework (not TF Lite) packaged for iOS that we use with the TensorFlow backend.
Unofficial build of TensorFlow in a self-contained CocoaPod that we use with the Tensor/IO's TensorFlow backend. Vends the tensorflow, protobuf, and nysnc static libraries and all headers required to perform inference and training on device.
Unofficial build of the full libtensorflow-core.a (not TF Lite) and required dependencies for Android, wrapped by the JNI, and packaged as a distributable Android library. Full support for inference and training on device.