HDR - MAXCLL AND MAXFALL LIMITS
Embedding static metadata in HDR10 videos offers several advantages, but also disadvantages. Static metadata such as MaxCLL and MaxFALL can in some cases even degrade the image if you don't pay attention to certain points. One reason for this are so-called outliers. If certain areas or individual pixels go beyond the limits, the screen naturally interprets these areas (pixels) and therefore misinterprets the image. As a result, the picture is darker and weaker than "planned".
When analyzing the metadata, such outliers are often not recognized correctly. The reasons for this are very different. Today we look at one of the possibilities with the outlier.
If so-called filter overshoots (overshoots) occur, artefacts can occur very quickly. The cause lies in the non-linear response of the electro-optical transfer function (EOTF), which produces outliers in the perceptual quantization (PQ).
The various filters and resizing ratios have different overshoot behavior. Some resizing ratios can produce a deviation of +2% to +10%. However, linear and cubic filters in particular hardly produce such outliers. Some filters do not have such desirable frequency response characteristics, and therefore often cause such outliers.
If lossy mezzanine compression is used, such as with JPEG 2000 and ProRes, outliers are sometimes generated even at high data rates. It is precisely such lossy compressions that can lead to small changes in the non-linear PQ values, resulting in pixels that exceed the limits (MaxCLL and MaxFALL). The fault lies with the decoding algorithms, which produce precisely these outlier pixels through decoding. And that's exactly why HDR should use mathematically lossless compression!
If you work in non-colormanaged mode (DaVinci YRGB) and set the resize filter to Custom, you can choose between 11 options.
In the RCM2 you can still choose between 6 different options. Here it is advisable to select the "Linear - Tone Mapped" mode to minimize the probability of generating outliers.
In our HDR courses we always point out the necessary head space and recommend not "pulling up" the specular highlights to 1000 nits, leaving enough space for outliers. We also recommend certain values for certain scenarios and of course also explain why you shouldn't crash in HDR, for example, and where skin tones should be set.
Finally, you should definitely look at the metadata document at the end as Dolby Vision expressly points out. But unfortunately such information often does not reach normal users. That is why we will be offering an extra slot in our next HDR master classes, especially for such pitfalls.