JVET-AN0052 [AHG17] [ Performance of GOP based RPR under CfE test conditions [Y. Sun, Y. Zhao, J. Sauer, J. Pardo, P. Jia, T. Solovyev, E. Alshina (Huawei)]

This document reports test results for GOP based RPR enabled in VTM software under CfE test conditions. Objective metrics results as well as visual examples are provided. Video examples are attached as part of the package.

It was requested in the presentation of the submission to re-issue the CfE with anchors using RPR, and evidence would be questionable as this had not been used.

Other experts disagreed with the statement that there would be no evidence, as various submissions did not use it.

It was commented that GOP-based RPR is a non-normative tool, and the upsampling step is not normative.

Several experts disagreed on using anchors with RPR, others agreed.

One expert commented that for CTC RPR should not be used, and better look into rate points where RPR does not give benefit.

For a CfP, requiring submission of higher rate point(s) would be useful, not to be investigated visually.

It was suggested that in a CfP two VTM encodings (one using RPR, one without RPR) could be used. As the anchor, the one without RPR should be defined, which marks the lowest quality. This was agreed.

It was commented that in typical bit rate ladders at least over a sequence at a given target rate is kept constant, whereas RPR can change.

It was commented that RPR might have more impact in the encoder run time category, and also reduce the decoder run time (or energy consumption)

The set of test sequences may need some alignment