New VVC profiles targeting Feature Coding for Machines
PositiveArtificial Intelligence
- A new study has introduced three lightweight profiles for Versatile Video Coding (VVC) aimed at enhancing Feature Coding for Machines under the MPEG-AI standard. These profiles, named Fast, Faster, and Fastest, demonstrate significant improvements in compression efficiency and encoding speed, with the Fast profile achieving a 2.96% BD-Rate gain and a 21.8% reduction in encoding time.
- This development is significant as it optimizes the transmission of intermediate features in split inference systems, which could lead to more efficient machine learning applications and improved performance in downstream vision tasks, thereby advancing the field of artificial intelligence.
— via World Pulse Now AI Editorial System