Dynamic Range test (real one)

Status
Not open for further replies.
Jannard

Jannard

Red Leader
This test was done by Graeme Nattress at the RED factory today. Great care was taken to perform the test accurately. The data clearly demonstrates 11 1/3 stops of dynamic range... or greater than 66db. Just as advertised back at NAB 2006.

1166058035.jpg


A Stouffer T4110C calibrated 41 step Transparent Step Wedge Chart was shot at F4.0 on the RED Frankie prototype camera.

The RAW sensor data was taken into REDCINE Prototype, and demosaiced. At no time was any noise reduction or data manipulation applied.

The data analysis mode of the REDCINE Prototype was used to align 41 sample areas with the 41 steps of the wedge chart. Individually red, green and blue averages were calculated for each of the sample areas. Log base 2 was taken of each average.

The graph plots the log green averages for each sample area against the calibrated value for each wedge in the test chart.

Graeme Nattress
 
:dankk2: Jim -- useful/real(ly one) and fair(ly) productive indeed :thumbsup:

EDIT -- Poi Boy had the true 11 figure!
 
Last edited:
That is great ! mine goes to 11; well, actually 11 and a 1/3.
Hopefully this will quite some of the whiner, snobs at cml. On second thought, it won't but who cares, mine goes to 11.
Aloha
-A
 
Jannard said:
This test was done by Graeme Nattress at the RED factory today. Great care was taken to perform the test accurately. The data clearly demonstrates 11 1/3 stops of dynamic range... or greater than 66db. Just as advertised back at NAB 2006.



A Stouffer T4110C calibrated 41 step Transparent Step Wedge Chart was shot at F4.0 on the RED Frankie prototype camera.

The RAW sensor data was taken into REDCINE Prototype, and demosaiced. At no time was any noise reduction or data manipulation applied.

The data analysis mode of the REDCINE Prototype was used to align 41 sample areas with the 41 steps of the wedge chart. Individually red, green and blue averages were calculated for each of the sample areas. Log base 2 was taken of each average.

The graph plots the log green averages for each sample area against the calibrated value for each wedge in the test chart.

Graeme Nattress
Click to expand...


Well, thats fantastic. I have no idea what you are talking about, but I see over 11 stops and I understand that are gud.

Out of curiosity - how does that compare to modern film neg?

In English, please :)
 
This is GREAT, thank you guys so much for doing this test for us. Any chance of seeing the frame that the range was judged from?
 
Great news, thanks for posting this chart! I'm curious... since the T4110C wedge has 41 discrete steps, I would expect to see steps or individual points on the graph. I assume the continuous line shown is drawn between the specific density values on the wedge (nominally 1/3 step apart).

At the bottom left region of the plot, it appears the response from approximately 2.5 to 3.2 stops is quite flat (horizontal line). Should that region be included as part of the effective dynamic range?
 
The chart is calibrated in optical density, from which the horizontal axis is defined. Yes, it is a touch flat there, but when you analyse the data, there's still information there that hasn't quite been swamped by noise. The graph scale would have to be bigger to see that, and being log to match the power law nature of optical density doesn't help.

Remember this is our first such chart, and we'll be using the tools as an internal reference as we work on the sensor to ensure we deliver great characteristics.

We think this method of measuring dynamic range is so much less succeptable to error than the series of 1 stop charts, which has a large opportunity for human error in both setting up of the tests and analysis. Our analysis is performed on the raw demosaiced data, in a very controlled environment, with no use of external software to measure an image, where you're never quite certain how the numbers are created, which makes it repeatable and accurate.

Graeme
 
Last edited:
This is very encouraging and much better than previous results. Would anyone at Red mind repeating the same test using another camera that the outside world can use as a reference? Any other camera?

Alternately, please do something like DPReview does when they test dynamic range using a Stouffer wedge. They post images and give more info than just a simple graph.

I do have a serious question, though, for future cameras: is there anything stopping you from putting little dots of ND paint on every fourth sensor or so and then using info from those darker pixels when the highlights start clipping? You could afford to give up some resolution for this because you have so much to start with.

I'd much rather have a 3K Red with great dynamic range than a 4.5K Red with only average dynamic range. I know I'm going to get mocked for this and I know the Fuji SR high-dynamic range system didn't do that well in still cameras. Still, dynamic range is so important for moving pictures.

Thanks!

Bruce
 
I don't know exactly what it all means, but I also know that 11 stops is very, very good.

I do know this.... You'll have to pry my reservation spot from my cold, dead hands if you want it.
 
How much of the dynamic range is useful?

From the Origin images up on CML (well practically every camera on the comparison tests there), the highlights tend to exhibit color shifting. Is there some metric/test where this can be evaluated? Dalsa is pretty open about their camera doing 10-12 stops, since those last two stops exhibit color shifting (my understanding anyways).

Perhaps putting error bars on there would be helpful, as it would indicate the amount of noise. And presumably it's possible to tradeoff noise versus grayscale tracking (i.e. apply processing to achieve good grayscale tracking; this can magnify noise at certain points of the slope).
 
Last edited:
Glenn, the test shows very good linearity. Luma linearity is even better that what you see in that chart. You'll get a colour shift in highlights when one channel clips before the other. The un-clipped channels still have data, and will show some detail, albeit in the wrong colour. With the settings we used for the chart as shown, the three channels all clipped at the same point (or very, very close) so there'd be no colour shift in the highlights. That means that all the dynamic range you see would be useful, to whatever you determine is your tolerable noise floor.

Graeme
 
bruceallen said:
This is very encouraging and much better than previous results. Would anyone at Red mind repeating the same test using another camera that the outside world can use as a reference? Any other camera?

Alternately, please do something like DPReview does when they test dynamic range using a Stouffer wedge. They post images and give more info than just a simple graph.

I do have a serious
Bruce
Click to expand...

good points Bruce. the DPreview does use the same wedge chart.. its a far superior way to judge latitude on a digital imager with within a single frame. Its basically the same as shooting 41 single frames of a luminant with a stack of 1/3 stop NDs. As Adam Wilt stated, its a little more relivant.

The actual wedges dont do much though looking at them, as most displays cant display the range, and even DPreview pumps the brightness and pushes the wedges.. if you look at the 41step chart they show where the clipping happens, its actually just a generated wedge from the data collected form the chart.
 
Spot on Jarred, but much less time consuming and open to error than taking 41 shots! That would be painful. Also, having it all in one shot means it's a doddle to automate the pulling out of the step wedge values. The first graph I did by hand - took about 30 minutes, and the second graph was done with the automated software, and it took less time to make this graph, even including the time to write and debug the software.
 
Last edited:
1- Graeme: From the images I've seen, I think I'd have to agree- grayscale tracking/linearity is very good. But if you could measure for useful dynamic range, wouldn't the Red fare even better in comparison to the competition?

OTOH, people have different ideas about what they can play with. So perhaps post stills of a high dynamic range scene + test chart would let people draw their own results. Shooting a scene can also be helpful since the final product will be an image, not a set of numbers like S/N ratio. People don't watch movies with vectorscopes. :D

2- Sort of along the lines of what Anders is saying, would using color filters help avoid one channel clipping (or getting too non-linear) before the others?
 
As Jarred and Graeme have pointed out, the charts on DPReview.com are manufactured from the data. They are NOT the images of the chart shot by the camera. We used the same methodology that is done at DPReview. It is absolutely the best way to measure dynamic range. Additionally, the noise from our Mysterium sensor is extremely low compared to other digital cinema cameras and comparable to the best DSLRs. Add in the resolution and you can see why we are all so excited about this project.

Jim

1166070872.jpg
 
Glenn - so good to have such great a discussion! Pity I never get long enough to chat to you when we meet up.

1) Yes, and in other tests we've done, it can look better still. Shooting a scene is great, but what we've got here is repeatable and produces numbers, and graphs, so when we work on the sensor, we can compare our changes in a repeatable manner.

2) True, but you'd need a different filter for each light source. We can optimize for any light source in camera, or in software, or you can do so with filters. Filters reduce sensitivity though. We have lots of choices here.

Graeme
 
The requests for high dynamic range scenes keep coming in... as we have stated before, Frankie is opposed to being moved around too much. Spike will be completely mobile and is due to be put together sometime this month. It is fair to expect us to post much more footage once Spike is up and running.

Jim
 
Jim, in working with the RED footage, to me, how it behaves in Photoshop is exactly like an image from a high end DSLR. What's amazing, is that you can also see it in motion!

Graeme
 
Status
Not open for further replies.
Back
Top