Extrapolating images with Boundless
- 0 Collaborators
The project demonstrates how to optimize the Boundless model using TensorFlow Lite and use it for image extrapolation. ...learn more
Project status: Published/In Market
Internet of Things, Artificial Intelligence, Graphics and Media
Overview / Usage
Image extrapolation is an important topic of study specifically in areas like image editing, computer graphics, and computational photography. For realistic image extrapolation, the Boundless model was proposed.
Methodology / Approach
The original Boundless model is ~125 MB in size which makes it difficult for mobile application developers to use it inside mobile applications. So, first, I optimized the model using TensorFlow Lite, then added metadata to the TensorFlow Lite model for easier integration with Android.
I generated two TensorFlow Lite models -- one is ~14 MB and another one is ~7 MB. These smaller models make it possible to deploy them to mobile devices.
Technologies Used
* TensorFlow Lite