Deep Learning on the Web: State-of-the-art Object Detection using Web-based Client-side Frameworks

Abstract

In the present paper we make a comparative study and evaluation of frameworks and libraries for deep learning purposes on the client-side, considering libraries such as TensorFlow.js, brain.js, Keras.js, ConvNet.js and others. It is examined how feasible and efficient it is to execute deep learning tasks, using client-side libraries and frameworks in contrast to the conventional approach. Moreover, we focus on the computer vision field of object detection and we examine the problem of object detection through different state-of-the-art approaches and object detectors. At the same time, we evaluate whether it is feasible and efficient to detect objects in the browser environment using a prototype implementation based on some of the libraries that are studied.