JVET-I0025 Clarifications on HDR Test of CfP on Video Compression beyond HEVC [P. Yin, W. Husak, T. Lu, F. Pu, T. Chen (Dolby), S. Iwamura (NHK)]

This contribution was discussed Saturday 20 January 2018 at 1120 (chaired by GJS & JRO).

This contribution requests a few clarifications regarding the High Dynamic Range Video tests described in JVET-H1002 “Joint Call for Proposals on Video Compression with Capability beyond HEVC”. The clarification relates to two aspects: 1) the input video format to the HDR display for coded test material submission; 2) reference software and configuration to be used to compute HDR objective metrics.

One issue raised is the conversion of the 4:2:0 YUV 10 bit decoded video (obtained from proponents) to the format needed for the display (e.g., possibly 4:2:2 YUV or 4:4:4). It was remarked that it would be desirable to have information about the exact process to be applied, so that a proponent could replicate the subjective display conditions. This could apply not only to the HDR case, but also to the SDR case as well. One test sequence ("Cosmos") was noted to be 1920x856 rather than 1920x1080, so part of that conversion process could include a resolution conversion process, if necessary.

• These aspects were further discussed on Monday at 1700 (chaired by GJS & JRO). The test coordinator said the active area of the Cosmos sequence will be vertically centered with black bars above and below (value 0, not 16). It was remarked that the BVM300 monitor does not accept 4:2:0 10 bit input, so conversion to 4:2:2 will be necessary. According to the NHK representative, the conversion to 4:2:2 that has been used in other recent viewing in JCT-VC applies 2×vertical chroma sample replication. Side activity (coord by V. Baroncini) was asked to meet in the test room @1900 to clarify other testing issues. In further discussion on Wednesday at 1215 (chaired by GJS & JRO) it was said that the test lab in charge has been using a hardware converter in which the conversion method was not currently known.
• These aspects were further discussed on Thursday at 1600 (chaired by GJS & JRO). It was noted that the intent of the test chair is to use the AJA HA5-4K device to interface between the 4:2:0 HDMI 2.0 output of the playback PC and the 4:2:2 3G-SDIx4 input of the BVM X300, and that this device will perform the needed conversion from 4:2:0 to 4:2:2. While this is the intent, it is recognized that the test chair and/or test facility may use an alternative device or configuration if determined to be needed while performing the test.

The BVM X300 reportedly supports two HLG modes: HLG SG 1.2 (HDR) and HLG SG Variable (HDR) (default system gamma value is 1.2). Since HLG SG Variable (HDR) is reportedly aligned with ITU-R BT 2100, it is suggested to use HLG SG Variable mode for the HDR HLG content test. It was verbally suggested that the system gamma should be set to 1.2 for the variable mode, and it was remarked that the variable mode with system gamma set to 1.2 was used in the evaluation of test sequences for selection. JVET response: This was agreed (using the variable mode with the system gamma set to 1.2).

Information about the version of HDRtools used for computing wPSNR was requested. This metric was not supported in HDRtools v0.15, which is the version of HDRtools that is prescribed for use in the CfP. The CfP says that the implementation in the JEM software is to be used (although it would be difficult for proponents to use this, since it is built into the JEM and that is not a codebase that proponents would be expected to be using). The contributor suggested using HDRtools v0.17. It was remarked that any difference between what is in the JEM software and the implementation in HDRtools v0.17 is negligible. The measurements used for the anchors were as specified in the CfP. JVET response Proponents may use either the v0.17 HDRtools implementation or the JEM implementation as they so choose, since the difference seems negligible. This conclusion was confirmed on Monday 1730. It was also noted that for some test sequences the provided anchor wPSNR measurements were computed using implementations of the metric other than the JEM encoder; the difference in all cases was in the fourth digit past the decimal point, so there was no apparent need to try to change them. This was agreed to be noted in the clarification document or the response template.

The contribution suggested that for the deltaE100 and PSNR-L100 metrics, the peak luminance should be set to 1,000 cd/m² for HDR-A (HLG) and 10,000 cd/m² for HDR-B (PQ). It was noted that the CfP says to use 10,000 cd/m² but also says that configuration files would be provided, and the provided configuration files for the conversion process used 1,000 cd/m² for the HLG content. Thus the suggestion from the contributor is in line with the config files that have been distributed. JVET response Agreed as suggested. If a proponent uses a different convention, the difference would be obvious and easy to correct by simply adding 10 to all provided measurements. This was confirmed in a follow-up discussion on Wednesday 24 Jan. at 12:20. A note about this was included in the clarification document.