Fast Text Detection

Overview

Our project is to create a fast text detection solution for video games, to help in the video compression and processing stage for cloud gaming. Using any approach available to us, with the requirement that we hit a runtime of less than 1 millisecond for a single frame of video, we are exploring this space in multiple ways, including Machine Learning. Often when live streaming anything across an internet connection, fine details can get lost in compression. Our text detection will inform the video compression server side to perform less compression on the areas that contain text. With a runtime of less than 1 millisecond per frame, the impact on performance is negligible, while the end user gets the benefits of a clearer image without knowing the system exists. In terms of video games, where text often holds important information for the user, this is critical to provide the best cloud gaming experience possible, as close to playing on real hardware as possible.​

Community Benefit

For the server side, our fast text detection will have negligible performance impact, while its output will help improve server-side video compression. For the end user, they will not even know this system exists, while getting an improved cloud gaming experience. Both sides of the cloud gaming service benefit, with little to no drawbacks.​

Team Members

• Jared​ Klein​
• Leonardo Martinez
• ​Nelson​ Mendez​
• Leonardo​ Osorio​
• Mykyta Palamarchuk

Sponsor

Intel Corporation