JVET-AM0234 [AHG17] Overview of the 3GPP codec testing in TR 26.955 [R. Mekuria, E. Alshina (Huawei), J. Lemotheux (Orange)]
This document provides a very brief overview of the 3GPP SA4 codec testing. This document targets assisting improvement of test conditions for potential next generation video codec development in JVET. This document just a very brief summary of materials made to the publicly available by 3GPP, deeper analysis is expected in a future.Presented Wed 2 July afternoon.
Services considered in TR 26.955 are
• 5G Media Streaming
o 5GMS Downlink Streaming
o 5GMS Uplink Streaming
o 5GMS Television
o 5GMS Downlink 360 Virtual Reality
• Multimedia Telephony Services over IMS
• Messaging Services
• Screen Content Coding
Codec support in 3GPP SA4 services is as follows:
a) HEVC/H.265 and AVC/H.264 have been recommended and supported in different services.
b) Also new codecs AV1, VVC and EVC were evaluated and considered for adoption (but none of them recommended).
The following quality metrics were used for in TR 26.955:
• SDR
o Per component PSNR: PSNR(Y), PSNR(U), PSNR(V) (if bitDepth is set to 10, then maxValue is set to 1020 – same as in JVET)
o Average PSNR over all colour components: PSNR = (6*PSNR(Y) + PSNR(U) + PSNR(V))/ 8
o Multi-Scale Structural Similarity Metric - MS-SSIM (there is minor difference in implementation compared to one used in JVET)
o Video Multimethod Assessment Fusion - VMAF. For VMAF computation the associated software implementation [5] is used.
• HDR
o Per component PSNR: PSNR(Y), PSNR(U), PSNR(V) (if bitDepth is set to 10, then maxValue is set to 1020 – same as in JVET)
o Average PSNR over all colour components: PSNR = (6*PSNR(Y) + PSNR(U) + PSNR(V))/ 8
o Per component weighted PSNR: wPSNR(Y), wPSNR(U), wPSNR(V)
o Average wPSNR over all colour components
o PSNRL100 (computed in 32-bit EXR)
o DE100 (computed in 32-bit EXR)
It was commented that it is valuable if different standardization bodies use different sets of test sequnces.
It was further commented that sequences used in CfE were selected to be appropriate for viewing, and it would be desirable to add more material for CTC.
The whole set of Netflix had been investigated in JVET, and selection was made by good reasons, e.g. not being complex enough.
It was commented that it rather looks as if 3GPP would have copied from JVET conditions, and that perhaps the test were conducted less rigidly than usually in JVET.
Interesting information, experts are encouraged to study the 3GPP report, at this moment no need is seen to change the test conditions in particular for CfE.