HDR Masterclass Notes
by Ben Allan ACS CSI & The New Zealand Cinematographers Society
Photo by Jess Charlton
Overview
HDR is a structure for handling and taking advantage of the increasing brightness and contrast of modern screens. Contrast is an essential element of this as increasing brightness on its own will result in a flatter, more boring image. HDR in this context should also not be confused with HDR stills which combine several images at different exposures to create as single image that captures a broader dynamic range. HDR for moving pictures is driven by display technology.
The other essential element of HDR is higher bit depths, both in capture and in presentation. The graph below shows how much more information there is at higher bit depths because each additional “bit” doubles the amount of tonal subtlety available for each of the three colour channels.
8-bit is most of what we are used to seeing on SDR television, streaming, DVD, Blu-ray etc. It is also what most DSLR’s, Mirrorless and Video cameras record. The h.264 format is generally 8-bit, as is HDCAM, XDCAM etc.
10-bit is Cineon (& DPX) Film Scans, HDCAM-SR, Digital-Betacam and most flavours of ProRes. Most HDR is delivered in 10-bit and it is considered the minimum for HDR acquisition.
12-bit is most RAW formats, eg. ARRI-RAW, BMD-RAW, Canon-RAW, REDCODE-RAW etc.
ProRes 4444 and 4444-XQ can also be 12-bit. DCP Digital Cinema files are encoded and presented in 12-bit.
16-bit is the basis of the ACES format and is usually transported in EXR files. 16-bit is also used in some RAW formats such as Sony-RAW.
In theory all of these bit depths can record an image with a Gamma or Contrast curve that is either Scene-Referred such as LOG or LINEAR or as Display-Referred such as Rec.709 or PQ. Best practice for HDR is that capture, processing and mastering should be done with a Scene-Referred curve and only presentation should be Display-Referred. In practice, 16 bit is usually Linear.
Understanding NITS
Nits is a measure of the brightness of reflected light. It is an alternative term for the metric measurement of reflected light Candelas per m2. The imperial equivalent is Foot-Lamberts and the Incident equivalents of these are Lux and Foot-Candles.
1 Cd/m2 is exactly 1 Nit.
The official benchmark for the brightness of a cinema screen is 14 foot-lamberts which equates to 48 Nits. Traditional SDR TV is meant to be 100 Nits, although it is important to note that most modern SDR TV’s are significantly brighter than this. Nits values can be measured in the real world by using a spot meter capable of reading Cd/m2. These can also be used to read the exact brightness displayed on a screen.
In the image below, the light coloured wall on the right is reflecting 300 nits while the same indirect, ambient daylight is reflecting only 42 nits from the dark green door.
In contrast to that, this image in the direct afternoon light, reflected 15,000 nits off the shiny white boat while only 2,500 nits of the deep blue of the harbour.
The angle of the light relative to the viewer’s position also affects the amount of light reflected. This footpath from a backlit perspective, kicks off the object to reflect 7,000 nits.
But the same light on that same footpath from front-lit perspective, was reflecting a much lower 2,400 nits.
Working in HDR Doesn’t Change The Dynamic Range of The Camera
While we may chose to grade a project in HDR or even monitor on set in HDR, a professional digital cinema camera will always be recording the same dynamic range as it would for SDR.
HDR simply changes what we do with that dynamic range.
HDR gives us the tools to makes creative choices that take advantage of the greater subtlety between tones and the impact of greater extremes between light and darkness.
HDR Systems
HDR10 is the base standard for HDR. It uses the Rec.2020 colour space which can display 75% of the colours which are visible to the human eye, compared to Rec.709’s 35%. It combines this with a PQ (perceptual quantisation) Gamma curve and 10 bit delivery and can be encoded into codecs such as any ProRes flavour or h.265 (HEVC) for delivery. HDR10 is an open source standard that is freely available for the use of the industry both on the content production side and on the screen manufacturing side.
HDR10+ is the same as HDR10 but with the addition of Metadata to govern the downconversion to SDR along the same lines as Dolby Vision but in a freely available, open source form.
DOLBY VISION is the proprietary system designed by Dolby Laboratories. It has the greatest precision of the HDR systems but requires licensing from Dolby both in the mastering suite and in the TV manufacture. Part of the grading requirements for Dolby Vision is to monitor a calibrated SDR display and Dolby’s special algorithm will make a shot by short analysis of how it should be down-converted to SDR. This needs to be done after each shot is graded and then in a licensed Dolby Vision suite you can also override this and trim the settings. Dolby Vision is usually regarded as the best master format because it makes it relatively easy to get to any other HDR or SDR format from there. However, the analysis phase does take time for every grade version of every shot and even on an extremely fast system like the one that we were working on at Department of Post, this time needs to be factored in to the grading schedule.
HLG is Hybrid Log Gamma and this is a system designed by the BBC and NHK to provide a simplified path to HDR. It works with a 2.4 Gamma Curve like SDR in the mid-tones and shadows but a Logarithmic curve in the highlights. This allows the images to display well on all HDR and SDR displays meaning that it is possible to have one delivery format that works across all display devices. This makes it ideal for non-graded programmes such as sport and other live and fast turnaround TV. The downside is that it does not have the precision and controllability of the other. HDR formats.
Test Shoot & Grade
The test shoot in the Unitec Studio was done with an ARRI Alexa XT using Zeiss Supreme Primes and Schneider Platinum IRND’s where required. We worked to the camera’s base ISO of 800 and a target T-stop range of T2 to T2.8 and we deliberately pushed towards the edges of the camera’s dynamic range capacity. Thanks to Ben Rowsell & Teresa Bradley for their camera notes.
Lighting and exposure were set by light meter and checked on the Rec.709 monitors provided by Rebel Fleet. One monitor was displaying the ARRI Log-C image and the other with a standard ARRI Rec.709 LUT applied. Waveform monitors displayed both of these as well.
The camera was recording ARRI RAW in 2880×1620 @ 23.98 fps. The images below have been imported directly as ARRI RAW into DaVinci Resolve in ACES and then delivered in HLG to provide a reasonable approximation of what we saw in the grading theatre for whatever display you are looking at this on. No grading or other processing have been applied to these stills (ie.they are equivalent to a 1-light work-print).
In the first setup the 29mm lens was set to T2.8 and there was no ND, while the camera was set to 3200K. The light coming through the window above the sink was full-CTB and from the left hand side window a simulated sodium vapour street lamp. The light on her face is purely from the iPhone 11 Pro which is half a stop under exposed and balanced to daylight.
The light from the practical is warmer than tungsten, probably around 2500K and in this shot the light falling on her face is 3 stops under as an incident reading. The reflected spot reading from the lamp itself is roughly 4 stops over. This was on the 35mm with an ND0.3 (1 stop) to get the aperture to T2 in order to get slightly shallower depth of field.
The next frame is from a later section of the same setup.
The light coming through the window is exactly on key exposure by incident reading on her arm. Bear in mind that it is full daylight blue though.
The daylight sequence is balanced to 5600K, still at 800 ISO. This closeup setup was shot on the
50mm at T2.8 with no ND. The highlight side of her face was 3.5 stops over on incident reading and 5 stops over on spot reading. The shadow side of her face was 2 stops under on incident.
The wide shot is identical in all aspects except on the 25mm lens. The spot reading of the light on the wall behind her was around 6 stops over at its brightest. This is close to the edge but not exceeding the camera’s dynamic range. This equated to 900 nits in the Dolby Vision Resolve project which because it is a large bright area was only able to reach 450 nits on the LG C9 display. Even at this level it was difficult to discern much detail in the shadow side of her face which was displaying at 3.5 nits or 7 stops below the bright areas. While the screen was able to easily display this difference, there was general agreement in the room that our eyes were not able to see detail in these shadows when they were so close to a large area of only 7 stops brighter.
This is a major consideration for HDR work. The capacity for the formats and the screens is able to exceed what our human vision can process at once. The obvious answer of course, is to employ the same sort of dynamic range control either on set or in post that we would traditionally do with DSR TV or Cinema in order to make sure it was a smaller area of brightness on the wall.
This reinforces the idea that the basic rules of cinematography and grading do not change with HDR, just that we have more scope in how we execute them.
In this later frame of the same setup, her face is 1 stop under (incident) instead of 2 stops under but also because it is more dominant in frame the difference is more dramatic and we are able to see her face much more clearly.
The backdrop is 2.5 stops over on incident reading and also on reflected when taking a spot reading from the water.
Conclusions
HDR is coming and we need to be ready for it. We are currently near the beginning of the transitional time much like the transition from 4×3 to 16×9 TV. In the short term we will need to be very conscious of both formats when planning and finishing projects. Right now, most of the audience will be watching SDR but it is not likely to be long before most TV’s on the market will have some form of HDR. Projects which are now being finished in SDR will most likely be reliant on the TV for conversion to HDR and this will be of variable quality. Finishing now in HDR gives us the best chance that our work will be seen the way we intend in the future. With HDR it is
important to remember that we don’t need to use the full dynamic range – it’s up to our creative intent.
The full video can be seen here: https://vimeo.com/383655072