Back to Search Document details
36th Meeting: Kemer, TR, November 2024 2024-11-07 11:59
Results of expert viewing for FGC subjective test category 1
Abstract
This contribution presents dry-run results achieved from an expert viewing test for film grain conducted at the 36th JVET and 17th SC 29/AG 5 meetings in Sapporo. The test targets at determining the potential of the film grain synthesis (FGS) technology to mask encoding artifacts an improve the viewing experience, in correspondence with category 1 of JVET-AH2022 Draft plan for subjective quality testing of the FGC SEI message (update 2) [1]. A subset of 6 test sequences provided in JVET-AI0138 [2] was selected for the visual evaluation, taking into account the viewing feedback from the category 2 test from JVET-AI0337 [3]. The sequences cover film material as well as digitally captured content with different artificial grain. The setting included 4 TV (3 miniLED and 1 OLED) screens which were driven by a PC video player.
JVET-AJ0374 Results of expert viewing for FGC subjective test category 1 [M. Wien (RWTH), P. de Lagrange (InterDigital), A. Wieckowski (Fraunhofer HHI)]

This contribution presents dry-run results achieved from an expert viewing test for film grain conducted at the 36th JVET and 17th SC 29/AG 5 meetings in Sapporo. The test targets at determining the potential of the film grain synthesis (FGS) technology to mask encoding artefacts an improve the viewing experience, in correspondence with category 1 of JVET‑AH2022 Draft plan for subjective quality testing of the FGC SEI message (update 2). A subset of 6 test sequences provided in JVET‑AI0138 was selected for the visual evaluation, taking into account the viewing feedback from the category 2 test from JVET‑AI0337. The sequences cover film material as well as digitally captured content with different artificial grain. The setting included 4 TV (3 miniLED and 1 OLED) screens which were driven by a PC video player.

  • MOS range
    • For some sequences, the lowest tested rate already exhibits a mid-range MOS – no observations for lower range of the MOS scale can be made
    • This is deemed appropriate given the set goal of the test
  • Results
    • For all tested points the MOS of the FGS encoding is larger than the corresponding NonFGS version, in most cases significantly
    • For four of the tested sequences the highest FGS rate points matches the MOS the transparency point. For the other two sequences, the confidence interval overlaps the transparency points. The highest FGS points always has a significantly lower rate than the transparency point.
    • For the sequences MeridianSmoker1 and MeridianIntoCar1 the benefit of FGS is the lowest among the tested sequences. This might be attributed to the subtle grain present in the original sequences.
    • The highest gain is observed for the sequence MarketStreet1, for which the up to 10x bitrate reduction without quality degradation is measured.

It was agreed to issue this as a report, and to include the results of JVET-AI0311 / AG 5 N 128 (the latter is a more refined version) as well.

It was suggested that software used to generate the results could be attached to a future version of the TR on film grain.

AHG4: Test and training material (3)

Contributions in this area were discussed at 1150–1255 on Tuesday 5 Nov. 2024 (chaired by JRO and MW – joint meeting with AG 5).

Decisions
It was suggested that software used to generate the results could be attached to a future version of the TR on film grain.
Citation