Back to Search Document details
15th Meeting: Gothenburg, SE, July 2019 2019-09-11 18:57
Description of Core Experiment 1 (CE1): Reference picture resampling filters
Abstract
This document describes Core Experiment 1 (CE1) on reference picture resampling. The experiments are conducted after the 15th JVET meeting and the results, including cross-checks, will be reported to the 16th JVET meeting.
JVET-O2021 Description of Core Experiment 1 (CE 1): Reference picture resampling filters [D. Luo, V. Seregin, W. Wan]

Regarding 0.5 second switching points, it was agreed to round down the switching point frame number to be a multiple of 4 to align with the GOP structure in LDB. In terms of the number of pictures tested, it will be set equal to 5 seconds worth of test material without rounding.

Because the dimensions of coded pictures (full-resolution and downsampled) must be a multiple of 8, it was agreed that the CE coordinators and participants will fill out the detailed plan for what lower resolution to code for each of the different source resolutions and different resampling ratios in the CE description document.

For PSNR1 calculation, averaging PSNR values of different picture sizes may not be desirable. One way to address this is to accumulate the MSEs of all pictures and only convert to PSNR at the end of coding. It was agreed that the CE coordinators and participants will fill out the detailed plan for PSNR1 calculation in the CE description document.

Decisions
For PSNR1 calculation, averaging PSNR values of different picture sizes may not be desirable. One way to address this is to accumulate the MSEs of all pictures and only convert to PSNR at the end of coding. It was agreed that the CE coordinators and participants will fill out the detailed plan for PSNR1 calculation in the CE description document.
Citation