concept tensorflow.js model in category deep learning

This is an excerpt from Manning's book Deep Learning with JavaScript: Neural networks in TensorFlow.js. Login to get full access to this book.

The second reason is the ecosystem. Most JavaScript deep-learning libraries define their own unique API, whereas TensorFlow.js is tightly integrated with TensorFlow and Keras. You have a trained model from Python TensorFlow or Keras and want to use it in the browser? No problem. You have created a TensorFlow.js model in the browser and want to take it into Keras for access to faster accelerators such as Google TPUs? That works, too! Tight integration with non-JavaScript frameworks not only boosts interoperability but also makes it easier for developers to migrate between the worlds of programming languages and infrastructure stacks. For example, once you have mastered TensorFlow.js from reading this book, it will be smooth sailing if you want to start using Keras in Python. The reverse journey is as easy: someone with good knowledge of Keras should be able to learn TensorFlow.js quickly (assuming sufficient JavaScript skills). Last but not least, the popularity of TensorFlow.js and the strength of its community should not be overlooked. The developers of TensorFlow.js are committed to long-term maintenance and support of the library. From GitHub star and fork counts to number of external contributors, from the liveliness of the discussion to the number of questions and answers on Stack Overflow, TensorFlow.js is shadowed by none of the competing libraries.

How to deploy TensorFlow.js models to various platforms and environments, ranging from browser extensions to mobile apps, and from desktop apps to single-board computers

TensorFlow.js is compatible with Electron.js, as is demonstrated by the simple example in the tfjs-examples repository. This example, found in the electron directory, illustrates how to deploy a TensorFlow.js model for inference in an Electron.js-based desktop app. The app allows users to search the filesystem for image files that visually match one or more keywords (see the screenshot in figure 12.7). This search process involves applying a TensorFlow.js MobileNet model for inference on a directory of images.

Figure 12.7. A screenshot from the example Electron.js-based desktop application that utilizes a TensorFlow.js model, from tfjs-examples/electron

Despite its simplicity, this example app illustrates an important consideration in deploying TensorFlow.js models to Electron.js: the choice of the compute backend. An Electron.js application runs on a Node.js-based backend process as well as a Chromium-based frontend process. TensorFlow.js can run in either of those environments. As a result, the same model can run in either the application’s node-like backend process or the browser-like frontend process. In the case of backend deployment, the @tensorflow/tfjs-node package is used, while the @tensorflow/tfjs package is used for the frontend case (figure 12.8). A check box in the example application’s GUI allows you to switch between the backend and frontend inference modes (figure 12.7), although in an actual application powered by Electron.js and TensorFlow.js, you would normally decide on one environment for your model beforehand. We will next briefly discuss the pros and cons of the options.