Intelligent Mobile Projects with TensorFlow by Tang Jeff;

Author:Tang, Jeff;
Language: eng
Format: epub
Tags: COM004000 - COMPUTERS / Intelligence (AI) and Semantics, COM074000 - COMPUTERS / Hardware / Mobile Devices, COM060180 - COMPUTERS / Web / Web Services and APIs
Publisher: Packt Publishing, Limited
Published: 2018-05-18T10:50:30+00:00

Recognizing Drawing with CNN and LSTM

In the previous chapter, we saw the power of using a deep learning model that integrates CNN with LSTM RNN to generate a natural language description of an image. If deep learning-powered AI is like the new electricity, we certainly expect to see the application of such hybrid neural network models in many different areas. What's the opposite of a serious application such as image captioning? A fun drawing app such as Quick Draw (https://quickdraw.withgoogle.com, see https://quickdraw.withgoogle.com/data for fun sample data), which uses a model trained and based on 50 million drawings in 345 categories, and classifies new drawings into those categories, sounds like a good one. And there's an official TensorFlow tutorial (https://www.tensorflow.org/tutorials/recurrent_quickdraw) on how to build such a model to help us start quickly.

It turns out that the task of using the model built with this tutorial on iOS and Android apps offers a great opportunity to:

Strengthen our understanding of finding out the right input and output node names of a model so we can appropriately prepare the model for mobile apps

Use additional methods to fix new model loading and inference errors in iOS

Build, for the first time, a custom TensorFlow native library for Android to fix new model loading and prediction errors in Android

See more examples of how to feed a TensorFlow model with input in its expected format and get and process its output in iOS and Android

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.