Add cine EI to any DSLR for free

[Nicolas POISSON]

This is just the good old principle of exposure compensated LUTs on an external monitor. I guess most people here are familiar with it: create “Log to Rec709” LUTs that darken or lighten the image by 1 stop increment, send them to the monitor, and use them as an EI scale instead of the DSLR analogue ISO (which you leave at the base value).

The point here is to create LUTs using maths and DaVinci Resolve. There may be simpler ways to do this with CSTs, but this is a topic I know nothing about. And part of the fun was to do it myself as a pedagogic exercise.

What I tried before:

1- shoot a scene with a middle grey card in it. Set the exposure according to the manufacturer advice and keeping ISO at its base value. Then shoot 1 and 2 stops overexposed, 1 and 2 stops underexposed. I guess the best is to change shutter speed (same depth of field), but aperture does the trick. Import all the stuff in DVR. Add the “Log to Rec709” LUT as the last node. Then add a second node before, and adjust offset with SDR wheels, or global exposure with HDR wheels, so that middle grey goes back to the reference. HDR tools work better, but both are far from perfect.

2- same as before, but fine-tune using whatever colour-correction tools you want (especially contrast, saturation, and the curve tool), so that compensated images visually match the normally exposed one. This is a trial-and-error method that leads to better results. But not very scientific.

After that, sandwich the colour correction node together with the “log to rec709” in a single LUT (one for each compensated exposure), send them to the monitor, and use them instead of the DSLR analogue ISO. You get Sony’s “Cine EI” feature on any DSLR. This works well for mid-tones, and this may be enough on set. But this can be improved.

LOG profiles are documented. Manufacturers give the formulas to switch from real world scene reflection to log coded values, and the other way around. Thus:

- starting from a recorded log signal, we can find what was the reflection

- compute what would have been the scene reflection if the sensor had been exposed 1 or 2 stops higher/lower

- from that simulated exposures, find what would have been the log codes for these reflection.

We can compute the curves that make a 1-2 stop over/underexposure look like if it was correctly exposed. I used Scilab to program this, but this can be done in any spreadsheet program.

slog3.sci

Here are the curves for SLog3:

Note: these curves are log space to log space, and SLog3 does not go below code 95 (10bit scale). The curve shape below 95 does not matter.

As the compensation curve will be drawn manually in Resolve, it helps to determine the points corresponding to the grid of the curve tool (25 intervals ranging from 0 to 100% Luma). Here it is for SLog3:

 

A few more notes:
- the curve tool in DVR uses splines, which do not allow for sharp edges. It may be better to have the lowest point at coordinates [0, 95] rather than [0, 0].
- upward compensation leads to values above 1023 (above 25th grid line). The best way to handle this is probably to draw manually some kind of highlight roll-off.
- From what I tested, 5-8 points are enough to define the curve (example below for Fuji F-log)

This gives a much better base than playing with offset or HDR exposure. This is still not perfect and I had to bring up lowlight saturation a bit for the negative compensation LUTs, and bring it down for positive compensation LUTs. But I could get good similarity across LUTs tweaking very few controls. Here are the results on Fuji F-log, starting from +2 stops compensation (underexposed sensor) to -2 stops (overexposed sensor). The 3rd image is the reference, exposed normally.

+2 stops (sensor underexposed by 2 stops)

+1 stop

Standard exposure:

-1 stop

-2 stops

The curtains lost all details in the 2 stops underexposed shot. Even with “only” one stop underexposure, there are quantization problems near the gray card, where the magenta turns into brown. My understanding is that my camera’s Dynamic Range (Fuji X-T30) is centred at +/-5,5 stops around middle gray at standard exposure. Underexposing the sensor by one stop leads to only 4,5 stops below, which I expect to be too low for many situations.

Highlights do clip in the 2 stops overexposed shot, visible in the specular reflection on the tiny aluminium PAR spot. However this scene does not contain much highlights, thus it does not appear that terrible to me. Following other tests I made, the most common exposure I would use is overexposing the sensor by 1 stop, which sets the DR at +4,5 / -6,5 stops around middle grey. Comparing to other manufacturers, this asymmetry toward low light seems common. Depending on the scene, overexposing by 2 stops might even be acceptable. Thus, I could think it as “the nominal exposure could be 1 stop above what Fuji says”.

Final reminder: write it down somewhere and tell the colorist. The meta-data claims ISO 640 whatever the LUT.

 

Edited April 7, 2024 by Nicolas POISSON