It’s definitely advisable to embed static metadata in your HDR10 videos, but doing so can actually end up making the picture worse on the viewing end if care is not taken to avoid outlier pixels – pixels that are unintended that fool the consumer’s display, resulting in a dimmer image.
Why don’t current methods for calculating static metadata disregard outliers?
“The industry was introduced to the content-related high dynamic range (HDR) static metadata items of MaxCLL and MaxFALL via the CTA 861.3 video interface standard, which was initiated in response to a request from the Blu-ray Disc Association (BDA)… When MaxCLL and MaxFALL static content metadata were invented by the BDA as part of the UHD Blu-ray format development process in 2014, the only HDR content that existed at the time were short test clips demonstrating the new format’s potential capabilities. These test clips did not have outliers. The industry only became aware of the potential existence of outliers in HDR content after real production and distribution started on a broader scale.”
Some causes of outliers
Outliers From Resize Filter Overshoot
“Outliers can be introduced in HDR content from resize filter overshoots… Filter overshoots are common artifacts of resize filters that can generate outliers due to the highly nonlinear response of the perceptual quantization (PQ) electro-optical transfer function (EOTF)… Different filters and different resizing ratios have different overshoot behavior. Common filters that do not generate overshoots (like linear or cubic) do not generally have as desirable frequency response characteristics as common filters that do generate overshoots (like Lanzcos).”
Commonly used resize filters can generate overshoot values of from +2% to +10%!
“At 1000 nits, a 10% filter overshoot due to a resize in the PQ domain translates to a 1992 nits pixel value, while the same percentage overshoot due to resize in display linear translates to 1100 nits.”
Lossy mezzanine compression distortion
“Outliers can also be introduced by lossy mezzanine compression of distribution service masters (e.g., ProRes or JPEG 2000), even when using high data rates that result in very high visual quality levels. The lossy compression distortion can introduce small changes in PQ nonlinear values. As described in the preceding section about filter overshoot, small changes in bright PQ-encoded pixel values can lead to large changes in corresponding light levels, which can lead to unintended outliers in the decoded data. This issue can be avoided if content metadata values are computed on uncompressed imagery or with files that have been subjected to mathematically lossless compression rather than lossy compression.”
So, in the absence of a heat map, what can ordinary folks do to avoid overshoots? In addition to performing HDR10+ analysis on the original RAW files, we can also carefully choose the resizing methods in DaVinci Resolve. However, the number of options is mind-boggling! In the Master Settings panel alone, there are a dozen or more, including Bessel, Box, Catmul-Rom, Cubic, Gaussian, Lanczos, Mitchell, Nearest Neighbor, Quadratic, and Sinc. When the Resolve Color Management presets menu is set to Custom Settings, there are a half dozen choices, of which Linear – Tone Mapped is supposed to yield the best results with most HDR media. We can also look at the HDR10 metadata document when export is complete to ascertain whether MaxCLL values are consistent with what was indicated in the waveform monitor. Finally, it’s probably not a bad idea to avoid grading highlights all the way up to 1,000 nits just because you can!
Source: On the Calculation and Usage of HDR Static Content Metadata, Michael D. Smith, Michael Zink
The Daejeon Chronicles 
Leave a Reply