Editor: Xiaoxianyu Taozi
[ New Zhiyuan Introduction] Recently, the annual PyTorch developer conference was held. At the meeting, Meta released PyTorch Live, a set of tools that can provide Android and iOS users with artificial intelligence experience.
Recently, the annual PyTorch developer conference was held.
At the meeting, Meta (formerly known as Facebook) released PyTorch Live, a set of tools that can provide artificial intelligence experience for mobile users.
PyTorch Live supports an programming language JavaScript, which can develop applications for two mobile operating systems of Android and iOS, and can also provide services for customized machine learning models for the broader PyTorch community.
At present, PyTorch Live is open source, version number v0.1.1.
https://github.com/pytorch/live
PyTorch Live open source project includes the PyTorch Live command line interface (i.e. torchlive-CLI), a React Native package that relies on the PyTorch Mobile library for inference on mobile devices, and an React Native template, as well as some examples that can be deployed on mobile devices.
PyTorch official website claims that PyTorch Live is a very easy-to-use tool library. Using PyTorch Live can help developers successfully build a mobile machine learning demonstration app within a few minutes.
PyTorch in recent years
In January 2017, Meta released an open source machine learning library based on Torch PyTorch.
Since 2015, TensorFlow has been the upper hand, but PyTorch has gradually heated up after its release and has been rapidly popularized in the developer community.
In recent years, PyTorch has become one of the fastest growing open source projects in the field of machine learning. Meta revealed that in 2019, the number of contributors on the platform increased by more than 50% year-on-year, up nearly 1,200 people.
Based on PyTorch Mobile in PyTorch, PyTorch Live allows developers to complete the entire process from training models to deploying models in the PyTorch ecosystem, and it also provides a React Native library that can be used to create visual user interfaces.
PyTorch Mobile was released in October 2019. Before that, Meta released Caffe2go, a new framework based on Caffe2 machine learning framework and is specially optimized for mobile CPUs and GPUs.
It is worth mentioning that no matter whether developers want to run PyTorch Mobile on mobile devices or other edge devices, it is OK. Even PyTorch Mobile can run on the server. "If you want to show a model running on Android and iOS mobile, it takes several days to configure the project and build the user interface. With PyTorch Live, development costs are reduced by half, and you don't need to have much experience in Android and iOS development."
PyTorch Live built-in tool
Currently, the beta beta version of PyTorch Live only supports Apple macOS operating system. However, support for Windows and Linux systems will also be available soon.
PyTorch Live has several dependency packages that need to be installed first.
One is Node.js. If Node.js is not installed yet, you can download it from the official website of Node.js or install it through Homebrew (i.e. brew install node).
The other is Xcode. If you want to run the PyTorch Live app on an iOS emulator or iOS device, you need to install Xcode from the Apple App Store. PyTorch Live currently supports Xcode 12.5 or higher.
During the development process, configuring the environment may be both cumbersome and difficult.
PyTorch Live provides an installation routine that automatically installs required dependencies. It will automatically try to install the following libraries and tools:
- Homebrew
- Android Command Line Tools
- Yarn
- Watchman
- Android Command Line Tools
- Android SDK
- Android SDK Manager
- Android Virtual Device Manager
- Android Emulator
- Android Emulator
- CocoaPods
PyTorch Live The toolkit provides a command line interface (CLI) and a data processing API.
CLI enables developers to build a mobile development environment and guides developers to build mobile application projects.
As for the data processing API, it integrates custom models in the PyTorch Live API, which can be built into mobile applications for Android and iOS.
Currently, PyTorch Live already supports functions:
. Image classification
Classify photos in the album
Real-time classification
1. Object detection
. Handwritten numerical recognition
Users can also customize some other small functions by themselves.
In the future, Meta plans to allow the community to discover and share PyTorch models and demonstrations through PyTorch Live, and will also provide a customizable data processing API and machine learning tools that support audio and video data. "It is our original intention to make it easier for developers to develop mobile applications and show the machine learning model to the community. At the same time, it is also an opportunity to share and use models by building a thriving community of researchers and mobile developers, and they can use ideas and communicate ideas." The task of
PyTorch is to accelerate the process from research prototypes to production deployment. With the continuous development of the mobile machine learning ecosystem, the release of PyTorch Live is more important than ever.
Therefore, some netizens call PyTorch Live "Game" changer! ”
Reference:
https://venturebeat.com/2021/12/01/meta-launches-pytorch-live-a-set-of-tools-for-building-ai-powered-mobile-experiences/
https://pytorch.org/live/docs/tutorials/get-started
