Pictures Decoding Time Estimation for Low-Power VVC Software Decoding

This paper addresses the issue of decoding time estimation aiming to reduce the power consumption of software decoding via Dynamic Voltage and Frequency Scaling (DVFS) techniques. Some high level syntax elements information is extracted alongside with each picture decoding time measured from a real-time software video decoder. Relationships between these syntax elements and decoding time are analyzed to select appropriate parameters for regression methods. Finally various regression methods are compared with regards to practical inference time and estimation accuracy required by a theoretical VVC Video decoding DVFS setup.