JVET-AG0116 AHG12: GOP-based RPR encoder control for ECM [K. Andersson, J. Ström, P. Wennersten, R. Yu, W. Ahmad (Ericsson)]
GOP-based RPR encoder control has been available for VTM since VTM-19.0. It selects coding in a reduced resolution when the PSNR impact of re-scaling from source resolution to reduced resolution is expected to be limited. ECM does not have such functionality. In JVET-AF0058 the GOP-based RPR encoder control was implemented in ECM and was demonstrated to show benefit for high QPs, but that contribution lacked results for CTC. The present contribution includes results on CTC.
Objective performance (BDR) is reported as:
ECM-11.0 with GOP-based RPR versus ECM-11.0:
- RA Y: -0,06%, U: 0,75%, V: 0,71%
- LDB Y: 0,01%, U: 0,09%, V: 0,03%
It is also asserted that subjective quality can be improved by the proposal.
It is proposed to add the GOP-based RPR encoder control from VTM to ECM to enable further experimentation of RPR functionality in ECM for improved coding efficiency.
In v2 results based on MS-SSIM were added.
- RA -0,29% 0,22% 0,05%
- LDB: -0,05% -0,03% -0,06%
The comparison performed in this contribution uses the normal CTC (rather than RPR CTC) as the anchor. The goal is to show that for 4K sequences, by switching the resolution on a GOP basis, a small luma performance gain can be achieved (though chroma performance loss is observed).
The proposed method is encoder only.
The proponent reported subjective improvement in offline viewing. No examples were provided in the contribution.
The cross checker suggests that the proposed change aligns the RPR encoder of ECM (no GOP-based RPR) and VTM (GOP-based RPR).
The proponent reported that this is the only difference between RPR encoder of ECM and VTM.
It was commented that such modifications are not the core activities for the ECM exploration.
Decision (SW): Adopt JVET-AG0116.
JVET-AG0116 AHG12: GOP-based RPR encoder control for ECM [K. Andersson, J. Ström, P. Wennersten, R. Yu, W. Ahmad (Ericsson)]
GOP-based RPR encoder control has been available for VTM since VTM-19.0. It selects coding in a reduced resolution when the PSNR impact of re-scaling from source resolution to reduced resolution is expected to be limited. ECM does not have such functionality. In JVET-AF0058 the GOP-based RPR encoder control was implemented in ECM and was demonstrated to show benefit for high QPs, but that contribution lacked results for CTC. The present contribution includes results on CTC.
Objective performance (BDR) is reported as:
ECM-11.0 with GOP-based RPR versus ECM-11.0:
RA Y: -0,06%, U: 0,75%, V: 0,71%
LDB Y: 0,01%, U: 0,09%, V: 0,03%
It is also asserted that subjective quality can be improved by the proposal.
It is proposed to add the GOP-based RPR encoder control from VTM to ECM to enable further experimentation of RPR functionality in ECM for improved coding efficiency.
In v2 results based on MS-SSIM were added.
RA -0,29%0,22%0,05%
LDB:-0,05%-0,03%-0,06%
The comparison performed in this contribution uses the normal CTC (rather than RPR CTC) as the anchor. The goal is to show that for 4K sequences, by switching the resolution on a GOP basis, a small luma performance gain can be achieved (though chroma performance loss is observed).
The proposed method is encoder only.
The proponent reported subjective improvement in offline viewing. No examples were provided in the contribution.
The cross checker suggests that the proposed change aligns the RPR encoder of ECM (no GOP-based RPR) and VTM (GOP-based RPR).
The proponent reported that this is the only difference between RPR encoder of ECM and VTM.
It was commented that such modifications are not the core activities for the ECM exploration.
Decision (SW): Adopt JVET-AG0116.