Efficient measurement of quality at scale in Facebook video ecosystem

Regunathan, S.L.; Wang, Haixiong; Zhang, Yun; Liu, Yu; Wolstencroft, David; Reddy, Srinath; Stejerean, Cosmin; Gandhi, Sonal; Chen, Minchuan; Sethi, Pankaj; Puntambekar, Amit; Coward, M. P.; Katsavounidis, Ioannis

doi:10.1117/12.2569920

Cited by 7 publications

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To avoid skewing the BD-rate averages from the use of verylow or very-high bitrate points (which would not be used in practical streaming), we keep results within the range of 50% to 96% of each metric's range by limiting the BD-rate integration limits. In order to expand the narrow scoring range of SSIM, we rescale it according to the FB-MOS proposal [11], which was shown to appropriately scale SSIM to the entire [0,1] range.…”

Section: Resultsmentioning

confidence: 99%

Toward Generalized Psychovisual Preprocessing For Video Encoding

Chadha¹,

Anam²,

Treder³

et al. 2022

SMPTE Mot. Imag. J

View full text Add to dashboard Cite

Deep perceptual preprocessing has recently emerged as a new way to enable further bitrate savings across several generations of video encoders without breaking standards or requiring any changes in client devices. In this paper, we lay the foundations toward a generalized psychovisual preprocessing framework for video encoding and describe one of its promising instantiations that is practically deployable for video-on-demand, live, gaming and user-generated content. Results using state-of-the-art AVC, HEVC and VVC encoders show that average bitrate (BD-rate) gains of 11% to 17% are obtained over three state-of-the-art reference-based quality metrics (Netflix VMAF, SSIM and Apple AVQT), as well as the recently-proposed non-reference ITU-T p.1204 metric. The proposed framework on CPU is shown to be twice faster than x264 mediumpreset encoding. On GPU hardware, our approach achieves 714fps for 1080p video (below 2ms/frame), thereby enabling its use in very-low latency live video or game streaming applications.

show abstract

Section: Resultsmentioning

confidence: 99%

Toward Generalized Psychovisual Preprocessing For Video Encoding

Chadha¹,

Anam²,

Treder³

et al. 2022

SMPTE Mot. Imag. J

View full text Add to dashboard Cite

show abstract

“…• the crf ranges span from 18 to 42 (for x264 AVC) and from 22 to 63 for the svt-av1 AVC and vpxenc VP9; • the utilized presets are shown in Fig. 1 and span the bulk of the complexity-quality tuning on offer by each encoder; • convex-hull selection takes place for all encoders and the BDrates for all surviving points in the convex hull are measured using slope-based integration and the Netflix libvmaf BDrate calculator [14]; • execution time is measured using GNU parallel on an Intel CPU (in our case this was an Intel Xeon 8275CL 24-core CPU); • VMAF, VMAF-NEG, SSIM and PSNR are computed using the Netflix libvmaf library [14] and in our case SSIM is rescaled so that it is more aligned to MOS using the FB-MOS rescaling [18];…”

Section: Video Codingmentioning

confidence: 99%

Domain-Specific Fusion Of Objective Video Quality Metrics

Chadha¹,

Katsavounidis

Bhunia³

et al. 2022

Proceedings of the 30th ACM International Conference on Multimedia

View full text Add to dashboard Cite

iSIZE UK(a) PSNR (b) SSIM (c) VMAF-NEG (d) VMAF (e) Proposed P.910-MOS Figure 1: BD-rate (Bjontegaard Delta-rate) vs. runtime of video encoders when assessed in terms of: PSNR, SSIM, VMAF-NEG, VMAF and the P.910-MOS fused metric derived by our proposal. The utilized encoders (x264 AVC, vpxenc VP9, and svt-av1 AV1, with and without preprocessing) lead to different BD-rate results for each metric. Instead of ad-hoc averaging of BD-rates, we propose to consolidate this difference via domain-specific video quality metric fusion with limited subjective testing.

show abstract