Deep Learning-Based Video Coding (DLVC)
-
Overview
-
Video coding technology is the fundamental and indispensable part of digital audio/video systems. Several series of video coding standards have been crafted in the past three decades, and several new ones are in the way coming. IEEE 1857 is a working group for standardization of video coding as well as related technologies.
-
The Future Video Coding Study Group (FVC-SG), affiliated with IEEE 1857, devotes itself to the investigation of video coding techniques that are promising for incoming and future video coding standards. DLVC, short for Deep Learning-Based Video Coding, is one video coding reference software maintained by the FVC-SG.
-
DLVC features a set of deep learning-based coding techniques as well as new or improved non-learning-based coding techniques. The framework of DLVC is shown as follows. To the best of our knowledge, this is the first open-source video coding software that integrates deep learning-based techniques into the traditional hybrid coding framework. It is quite promising: DLVC version 1.0 achieves more than 40% bits saving than H.265/HEVC at the same quality level! But it is developing and far from being mature. So please join us in improving it.
-
Enjoy your hacking with DLVC!
-
DLVC Reference Software Version 1.1
-
Source code:
-
Executable:
DLVC Reference Software Version 1.0
-
Source code:
-
Executable:
Documents
-
The technology proposal can be previewed
here
.gif)
