MOVIELABS/DOLBY MEETING JUNE 19, 2013 SUMMARY: The meeting went until 11PM! Many topics were covered. I took extensive notes, which I condensed (believe it or not) to the below. There was a great deal of info so I wanted to make sure we had it documented. The topics covered are: El Dorado Demo at 4000 and 2000 nits Color volume gamut analysis of El Dorado and other clips Perceptual coding and PQ curves EDR consumer workflow Mapping EDR to lower dynamic range televisions Standardization Proposed path to affordable displays Workflow and production solutions EDR display manufacturers and NDA IMF Workflow Security issues General comments from the studio attendees Specifics follow below.
2 EL DORADO DEMO: They did show El Dorado at 2000 and 4000 nits-they said they had some issues on Monday and wanted to show it again. I found them quite different, actually. As you already know, it was graded at 4000, and the 2000 was created via an electronic mapping that a Dolby colorist used and then did some tweaks. My take was was that the 4000 had really good detail in backgrounds and very good differentiation in colors, specifically the different tones of red, but blues were also very distinct. I thought all of the color tones were very well represented, but some of the colors seemed "over the top", noticeably the rocks. Others noticed it too. Dolby said it was a characteristic of the monitor. Hmm. My take on the 2000 was that it looked very pale color-wise in comparison all around, though a little more natural in a few spots. They both had a LOT of judder, which they said was a timing issue in the monitor. In my opinion the judder was a real problem. No banding was perceived. Noted that Dolby Pulsar monitor uses P3 primaries, P65 White point POST DEMO DISCUSSION They then showed a gamut analysis of El Dorado at 4000 and some other content they had, which was very informative. Many pixels were out of the usual gamut range and evident in the plot. As a reference point, using overall10000 nits, in P3, blue is 800 nits. At 100 nits blue is 1 nit. Concept of color volume is key. Old films have 10 stops of dynamic range. Comparing an IP scan to negative scan reveals very different color volumes. In today s color timings, often, color is compromised to get more brightness. Pushing up the red allows more perceived brightness than just using the yellow. Pat Griffs notes that 4000 nits good but not all the way there. For example, the Arri camera has more information than can be shown at 4000 nits.
3 PERCEPTUAL CODING, PQ CURVES-Scott Miller Design color encoding to be tailored to human perception. With HDR, the cost is that you need more bits to represent it. So use PQ non-linear encoding to distribute the bits we currently have more efficiently than gamma encoding. To determine best coding parameters, utilize Just Noticeable Difference (JND) threshold. 1JND is at the edge of what can be perceived. They showed an experiment with 26 "patches" of data with different colors, against a grey background at different levels.1 to 500 nits. Plotted how people would perceive the differences at different light levels, colorimetry, bit rates, etc. At 12 bit XYZ the curves look good, the difference between 2K ans 10Knits is.3 JND for the middle color range. At 10 bit XYZ, the JND goes up quite a bit compared to 12 bits, and at 8 bits is quite high. The conclusion was that if XYZ was to be used it should be at least 12 bits. Science says that REC 2020 at 10000 nits has the best packing of data, but REC 2020 is not a future proof container. Wendy says that broadcasters would go for REC 2020 for the efficiency, and monitors would build to it. Says that cap-xyz is really good. Would CE manufacturers put chips in their sets that could do both? It was noted that true REC 2020 has laser primaries, NHK has a monitor with these primaries. In general, the use of more than 3 primaries is of interest. There are 4 primary OLED s today. Craig Todd attended a meeting in London with 85 broadcasters, they saw EDR on UHDTV. They noted that live action content does not go out of P3, but, they were very interested in the luminance factor. Dolby shot a lot of images in nature and they found very little outside P3. PQ is not standardized, it is Dolby IP. Dolby said that ITU was starting a standards effort. However, PQ would be licensed and not given free of IP. Howard Lukk was not happy with that.
4 EDR CONSUMER WORKFLOW The use of the HEVC/H.265 codec was discussed for the EDR application. Dolby is trying to use existing hardware-two 8 bit H.264 decoders to get the EDR signal for HD-upper layer and lower layer. Can do this on a tablet. This way cold deploy quickly. This also leaves the bottom layer as being backward compatible. For UHD, can use one H.265/HEVC and one H.264/AVC decoder. Dolby Composer adds these together into 12 bits. Output of composer cannot be shown directly on todays tv s, the content needs to be mapped Another concept that was discussed of using a 10 bit decoder capable of 60 fps at 24 fps so can push extra bits thru it. It was noted that 10 bit XYZ at 500 nits is worse than current BD. Lower luminance levels show a great increase in JND error. Dolby says to do XYZ, have to do 12 bits in order not to have this JND error in the lower luminances. Rec 2020 10 bit with PQ is a little better. The eye is sensitive between 100-500, above that, goes to logarithmic Going from gamma encoding to PQ encoding buys about 2 bits overall MAPPING EDR TO LOWER DYNAMIC RANGE TELEVISIONS Typical TV ranges from 350 to 1000 nits. 500 nits is often used as the average. Will have 2000 nit displays in 3 years. There was a long discussion of finding a method to map EDR onto lower dynamic range TV s and maintain hues and perceptual intent of peak values to average. This involves sending EDR DM metadata along with the EDR signal, possibly utilizing OTT. The mapping is created by taking a histogram of pixels in the EDR scene or still image and taking a metadata snapshot of it. The metadata is generated based on the artistic intent of the original piece. Can be edited by the colorist. Can view at several operating points and can modify the algorithm. The metadata is per scene, can switch on a frame CE vendors are talking about putting mapping into their displays for brightness and color volume to the TV s capability. Greg Gewickey said the demos show you can get 97% there by automation, but then want to do some tweaks. Wendy says they should work with the colorists to make sure the mapping works at a lot of different points. Said would use LCD versus laser display. Would not master on laser
5 and then do a trim pass on an LCD Dolby says you have to master at the widest possible range. Craig Todd saying that ITU said to master in highest quality, which was REC2020. Dolby blocked because they think we need to be ACES, openexr, etc. STANDARDIZATION-David Brooks EBU and SMPTE do not want constant luminance. Curves proposed to ITU were PQ REC 2020 at 10000 nits. Wendy says need to apply at PQ curve to cap-xyz at 12 bit. Dolby says they are working on a solution for projection that is very much more dynamic range. Show in 6 weeks. Current is 2000/1. They are talking about 10000/1. Not only brightness, but the black level as well. Some core changes would be needed to the MPEG codec. To what extent will what we are asking for effect the base codec or just a change to parameters? PROPOSED PATH TO AFFORDABLE DISPLAYS Dolby working on a monitor using quantum dots. In a typical HDR monitor, there are 18000 White LED s for the backlight. These LED s go off with no image. Replace these LED s with Blue LED s at $.05 each. This makes a blue backlight, need an optical stack sheet in front of the backlight to deal with this. This stack has the ability to give exact wavelengths. These quantum dots re-emit the color of choice-for example, red and green. This is then side lit. Saw a demo that looked pretty interesting of a prototype quantum dot monitor that was side lit against a broadcast monitor. Looked promising. It was noted that in an RGB system, blue is only used 15%, but those LED s are the most efficient OLED s are only good for 1000 nits for a couple of weeks
6 WORKFLOW AND PRODUCTION SOLUTIONS Noted that today s film stock has a very wide dynamic range, as good as digital. Scanners are getting better too. Howard mentioned the Panasonic organic CMOS with 26 stops. Need to preserve dynamic range thru the process. Raw footage has more range than the DI, and usually no one goes back to see that footage. By having the EDR displays, can see this up front. For ACES workflow, Dolby says they have a better IDT. Current ones throw away a lot of data. Having a better IDT means you don t have to do an initial color grade to get back to what you had. XYZ can be converted to ACES and back in floating point without loss. Even for DI, should do initial grade in HDR to see what you have, even though you need to go down from there for D-Cinema. In Post, Dolby CM (content mapping) tool would create metadata that describes each scene in terms of its dynamic range. Can see min and max in interior and exterior scenes. This is created algorithmically. Can override for 100 nit, for example, to crush blacks. Can copy and paste metadata when cutting back for the same scene. Metadata is frame based, can change it dynamically if needed when moving from dark to light for example. For editing, would we be able to do a CM proxy that would let the director see more in the Avid. Howard says doing a lot of on-set color correction, then make DNX-HD, straight to editing. No colorist Dolby said they are working with Film Light (baselight) and Resolve to get the CM tool into their toolkit. Would be an OpenFX plug-in to the baselight. The same Dolby software guy is working on this who did the auto 3D. Dolby wants feedback on which tools to support. Group felt Baselight, Resolve, Film Master, Misitika and Scratch VFX pipeline will be critical for the EDR pipeline. Nuke is a primary tool here Concept discussed of saving different versions of CM metadata for different light level media. Archival master must come from the EDR grade. Archive would be EDR master in half float? The metadata would be part of the elementary stream on a BD GG asked about Fragmented MP4 for OTT.
7 Some asked about watermarks in base content Wendy had a concern about putting the metadata directly into the stream early. IMF WORKFLOW Metadata track can be in the servicing master for IMF and played by the CPL. Metadata is XML, will be able to be in IMF OPL. Mezzanine file would be J2K, 12 bit EDR EDR DISPLAY MANUFACTURERS AND NDA Wendy asked about display manuf that can support HDR, Dolby says cannot say, they have an NDA. Wendy asked that these manufacturers talk to the studios. Hanno says can do thru Movielabs. We may be contacted. Dolby Also talked to the SOC manuf.by using dual AVC encoder approach, need to add very little real estate. SECURITY ISSUES Wendy putting together provisions for next gen hardware that will include security needs. Hanno says the layered approach Dolby proposed has issues with security Was noted that paytv people also interested. GENERAL COMMENTS FROM STUDIO ATTENDEES Wendy says HD is the current SD. Can t call HD a premium experience to be monetized. WB looking at EDR as the next thing, but it needs higher security. Hanno says they are going after the 85 UHD, support 10 bit from the beginning. Scalable approach-deliver 12 bits with a scalable codec. If you have 8 or 10 bit device, would only decode that much. Hanno saying BD wants to go EDR, just don t know how to handle the costs. Wants to only output EDR and figure it would work out. This conversion from EDR to standard has to be in the players up front so the studios don t have to supply more SKU s. Premium copy with UHD, HD and DST