Dynamic Range test (real one)

Status
Not open for further replies.
Lets get this all in perspective. I found the images David Stump produced very encouraging, I found them encouraging when I first downloaded them. I was under no impression that this was the best that RedOne is ever likely to produce. There have been many caveats issued enroute to all the images posted. I think it is clear that the sensor is uncharacterized and there have been improvements with the way the data from the sensor is interpreted, this was known and made clear when the images were posted.

Geoffs graph was interesting. I think it was a mistake not to make it clear on the graph that it was work in progress, but CML does assume some level of interpretation of information posted there, it is not intended for novices. Graemes work with this test is again very interesting but it has limited meaning because it isn't comparative and it too isn't using a production camera of course.

The CML graph didn't write off RedOne for me and this graph doesn't make me think RedOne is top of the pile. I would reiterate what I said on the "latest info" thread. I think it is important to bear in mind that this is all work in progress and nobody should be coming to any conclusions about RedOne at this stage. The only tests that really count are independent tests using the finished product and they can't happen until the product exists.

I don't think OTT fanboyism and rubbishing of the contributions over at CML is particularly helpful to the Red team or the dvxuser forum. I like the enthusiasm over here but lets not get over excited, Jim and the team have got a way to go before they are out of the woods. Lets enjoy being part of the development process and try to contribute helpfully, (that means both positively and negatively).

Sometimes it feels like it's Red versus the rest of the world, not because of the way red is attacked, but because of the speed with the which the drawbridge keeps being pulled up. Lets all just chill a bit and not be so keen to talk-the-talk all the time.

Martin
 
Hi Red Team,

I just wondered, a while ago you said you'd found a way to extract more lattitude from the sensor with some clever re-programming.... So my question is this, this new 11 1/3 stop test that you've achieved, is this with the new modifications made to the sensor or with the old set-up that David Stump used to get his 8 1/2 stops?

Have you gained more lattitude simply out of a new (more accurate) testing method, or did you actually increase the sensors DR after Stump's visit.
 
I think what we have is a more accurate method. Now we can start with more geeky changes to the sensor to see what they can do for us, knowing we have a repeatable reference, and can generate a new graph in a matter of minutes.

Graeme
 
Stephen Williams said:
I never stated that I could set a Cooke S4 lens exactly. I am sure I would be within 1/10th stop, as the wedges are 1/2 stop intervals, I don't see that as a major problem myself.
Click to expand...

The wedges are 1/3 stop intervals, but each wedge of the chart is calibrated to a much more precise level than 1/3 stop.

Especially as you get down to the dark regions, the slightest (and I do mean slightest) error gets magnified and you won't be able to see where there is any remaining detail in the blacks.

Graeme
 
I notice that Geoff over at CML has said when he does the next test series he will include the Stouffer T4110c test. So we should get to see some comparative results next year. Excellent.

Martin
 
Graeme_Nattress said:
I think what we have is a more accurate method. Now we can start with more geeky changes to the sensor to see what they can do for us, knowing we have a repeatable reference, and can generate a new graph in a matter of minutes.

Graeme
Click to expand...

So you're saying the old test of 8.5 stops was inaccurate and according to your methods it's 11.5 on the same footage? e.g if the milk footage was shot again it wouldn't look any different to what we have already seen in terms of DR.... I guess if David Stump came back he'd use the same method of testing as he did before and therefore get exactly the same results? Or are you asking him to come back once you've finished more sensor tweaking?

Either way if 11.5 stops is the starting point then thats pretty good!

What we need is to see all those other cameras measured with the same technique otherwise it's very hard to draw conclusions. (of course that's not Red's job, but is that info out there at all?)

BTW if any of this seems like a moan it isnt. You're doing an amazing job and I cant wait to get my hands on the camera. I just take great interest in the development of it all and how you're coming along.
 
The RGB Tiffs that David received were processed to generally look "nice" - the default raw conversion, which is not optimised to gain extra dynamic range in artificial tests. I think if I took David's frames and re-processed them, I'd be able to see the 11.5 or so, but that's a lot of frames to process and analyse, but with this new method, it can be done in minutes, and it's accurately repeatable.

Graeme
 
Hear, Hear Martin

Hear, Hear Martin

Martin hits some key points in his post about the vitriol the Red project has stirred up in the community and the need for some perspective at this stage of the development process. I also feel the Red Team's need to stand up for their project and the reservation holders who were getting pilloried by their peers over the 8 stop DR measurement. :violent5:

Ted refers to himself as the "leader of the rebellion" which implies an existing structure to rebel against. A lot of this brouhaha is inevitable. Now, those of us who are not actually on the Red Team can enjoy the holidays and look forward to 2007, which fanboys like me refer to as the "Year of the Red".

Blair S. Paulsen
RedOne #19
 
Graeme_Nattress said:
The wedges are 1/3 stop intervals, but each wedge of the chart is calibrated to a much more precise level than 1/3 stop.

Especially as you get down to the dark regions, the slightest (and I do mean slightest) error gets magnified and you won't be able to see where there is any remaining detail in the blacks.

Graeme
Click to expand...

Graeme,

With all due respect I was referring to David's Stumps tests. AFAIK they were 1/2 stop wedges, so I believe my comments still stand.

Best wishes,

Stephen
 
They were not un-processed at all. I don't have the frames and didn't do the processing, but looking at them, they were not optimum. Also, with Dave having stepped on the cable, we had a lot of bad noise in the image, which isn't normally there. I'd prefer to leave them as a historical artifact.

Yes, Dave's test were in .5 stops. Sorry, thought you were talking about the stepwedge chart.

Graeme
 
The defaults I put into the REDCINE proto were designed for a basic "nice" look. There's a fairly pretty looking contrast curve in there (that will crush highlights and shadows), and indeed, there was no knee processing of highlights at that time, and if I remember rightly, the gamma was not REC709 but a rough approximation. The proto changes rapidly around here. The current version is very different and more advanced.

So, if those images Dave took show 8.5 stops, that's because that's how my code processed them. If I'd set better defaults, they'd have looked better. My fault entirely.

Graeme
 
In review, the processing error was my fault... things are moving so fast around here that stuff like this happens. But I promise not to do it again.

Jim
 
IMHO, that's the results* that counts. If the numbers (resolution etc) standard is guaranteed...

(and it is the same poster who left the HVX purchase 'cause its lower CCD native resolution who is saying...)

EDIT -- * Imaging...feel of the image, that «basic "nice" look» as Graeme has been defending...and (above all) scoring.
 
Last edited:
alacritymedia said:
Martin hits some key points in his post about the vitriol the Red project has stirred up in the community and the need for some perspective at this stage of the development process. I also feel the Red Team's need to stand up for their project and the reservation holders who were getting pilloried by their peers over the 8 stop DR measurement.
RedOne #19
Click to expand...
The Red project itself has stirred up nothing.

A few big mouths on both sides of the fence are causing 90% of the bad blood. Geoff's chart is what it is and Red came back with better tests. Done. Red should be putting out better data or not putting out anything at all until a camera is ready. I saw the chart and was a little concerned but felt it was sort of irrelevent because the camera still has alot of development work to go. If anything, the chart simply brought "top end" issues to the front of the discussion which is a good thing.

Early yesterday morning, a mean spirited thread about this got started over at DVinfoNet and the mods quickly removed it.

Thanks DVinfo for taking the high road.
 
Last edited:
Sproketz said:
Early yesterday morning, a mean spirited thread about this got started over at DVinfoNet and the mods quickly removed it.

Thanks DVinfo for taking the high road.
Click to expand...
I think you're talking about of that thread opened by Robert Jackson, even if he's a fair guy and his intentions are just good ones.

If your concern is what I'm thinking: the energy spent on this coming from the RED team on your quoted «bad blood». I can just agree, even if I don't believe they care about such damage in any way other than as inform us. It's marketing too. Good marketing. And it runs for the truth sakes.
 
Thanks Graeme.

I had no idea that the frames David took away had been processed. This makes full sense now.

EDIT

Out of interest, has the 'milk' footage been converted using your old processing or the new, higher DR processing?
 
Last edited:
Nice results indeed. I find it kinda funny that you came up with this testing procedure, which has been used for scanner testing for decades :D

Some years ago I worked for a development company in the highend scanner market here in Austria.

When evaluating a scanner's performance we also used the Kodak grey wedge and let a little script extract the data to plot the latitude.
There are also several other testing methods, including the possibillity to plot the sharpness curve of the scanner lenses. We could also perform a spatial calibration to smooth out any imperfections in the lenses, glass plate or mechanical drive. That would go well par into the subpixels.

I think that kind of calibration could be done today to even out any chromatic aberrations and make cheap lenses more valuable.
 
This is a really educational thread. I love all of the technical discussion of stops of exposure, even if I don't totally understand it.
Can RED's end-users expect the capacity to tweak these settings individually, or will it be more of a Pana-style "cine-look" "xyz-look" settings arrangement? Would these settings be things we would even want to tweak, or is it totally unneccessary after the sensor is corrected, calibrated, et cetera?
And also - is this hard clipping we're talking about here, or does the RAW format of shooting allow for some adjustments in latitude through the capture system, like film? I seem to recall hearing something like that...
 
Status
Not open for further replies.
Back
Top