- Posts: 1073
- Joined: Sun Sep 20, 2015 12:11 am
- Location: Tallahassee, FL
Gene Kochanowsky wrote:John Brawley wrote:Uli Plank wrote:IMHO the best way to measure it is shooting a Xyla 21 from DSC.
Unfortunately that thing is expensive and needs a pitch black studio.
Yes.
It needs to be done properly. Not only that, but you'll still get differences in opinion on what's acceptable noise. Some will see 14 stops and some will see 12.
DR is a subjective measure. One mans noise is another mans shadow range.
I personally think that noise should also be evealuted in MOTION. Your perception of noise changes from a still frame to one that's moving but no one ever shoots DR charts like that either.
JB
John, What do you think of this experimental setup? It will be crude but I think it will be effective.
1. Using the following set of nd filters construct a test apparatus similar to the Xyla21
https://www.bhphotovideo.com/c/product/ ... y_Gel.html
2. Since I'll be using gel ND filters the chart will be illuminated from behind using a light constructed from one of these:
https://www.amazon.com/gp/product/B01DX ... UTF8&psc=1
3. Aluminum foil painted black will be used for the opaque parts of the apparatus.
4. I will setup the apparatus so that it will flip up and reveal the light so an image can be taken of the light intensities illuminating the gels.
5. Then an image will be taken of the apparatus back illuminated.
6. Steps 4 and 5 will be repeated a dozen times or so to take a statistical sample to examine variation in light values over the course of the test.
7. I will use a program that will allow me to review the frames and measure light intensities in bit values. (Not sure if Resolve has this functionality)
8. All this will be recorded in a spreadsheet for analysis.
9. Maybe I'll write up a paper and post it on the board.
If light uniformity is in question I could construct a slit in the light and move it across the apparatus to use the same light source for all stops.
Hard to answer from set on an iPhone.
It's not the way I do it but it seems fine if you want to emulate the test chart model.
I prefer to shoot a person in controlled lighting.
I put a range of lighting into the scene. Aim for say 5 stops.
I'll also put what I call an exposure ramp. Take a Fres type lamp and run it along a wall in the background. The point closest to the lamp will generally meter f22 on a spot and it peters off to nothing at the end of the beam.
Then I'll get a light trap. A light tight shoe box enclosed and painted black on the inside. Cut a small hole in one end and face to camera. This is your "true" black. This is an important reference for the noise floor. This black is always black no matter how far you lift it. It helps tell you when you've lifted the exposure too much.
THEN
You do bracketed exposure takes. Make what you decide is the zero "correct" exposure. Then do +1 stop, +2 etc for as much as you need or want. Then do -1 and -2 etc.
THEN
Grade your zero exposure how you like it. Distribute the 5 stops.
THEN try to match each exposure bracket to the same grade. By trying to recover the gradual under or overexposed images you get to see the DR clips (add the bracket to the different 5 stop scene exposure readings) plus you're also testing the recoverability of your bit deapth / codec combo.
Many times you can have an 8 bit codec that looks OK out of camera. But as soon as you try to reposition something that sits at 80% on the vector down to 35% you'll find it technically can do it but it looks like total crap. Plus this way you're testing a shot in motion.
So that's how manufactures can have high DR claims but it's all BS until you shoot in the real world and work with it.
By the way, I think the lessor known Arri test chart is better if that's what you want to replicate b
https://www.arri.com/arriajax?mod=produ ... roduct=263
JB