Building a Pokédex to recognize Pokémon in real time using TensorFlow and object recognition

Description

Back in the day, AI was a field dominated by people from academia or engineers from top companies. It required a very high level of mathematical understanding and very challenging technical training to be able to do anything with it. The recent popularization of AI, effective collaborative learning, the open-sourcing of bleeding-edge technology and tools, and the vast availability of high-quality data, along with cheap computing power, have opened the doors of innovation to the wider audience. Therefore, in recent years, its importance has greatly risen. TensorFlow enables us to reach that level of sophistication while leveraging technology and tools created by the very best, who have done a lot of heavy lifting for us, letting us just dive in and create something from the cool ideas.

Machine learning and object recognition have matured to the point that exciting applications are now possible. Anmol Jagetia demonstrates how to create a Pokédex that uses a camera phone to recognize the Pokémon it’s looking at in real time. You’ll explore core concepts of ML research, like gathering data, cleaning data, preparing a dataset, optimizing data to handle the problem, and then optimizing it to use as little computation as possible while still remaining robust enough to enable real-time predictions. You’ll learn how to reuse learned logic and models and use them over a set of similar problems and discover how to deploy your model to a mobile device, allowing a far more elegant solution to the problem than if it could just run it on a server.

Topics include:

• An introduction to Pokémon and TensorFlow
• A primer on basics of Python, the TensorFlow Object Recognition API, and object recognition techniques
• An explanation of the ML pipeline
• A short walkthrough of the entire process—how each step was done, the challenges faced, and the trade-offs made
• Demo of an already trained model and a working implementation