8 1/2 STOPS - latitude color grading chart

[boothba]

NOTE: 8 1/2 STOPS is the range of this test - not the range of RED's sensor. I would change the title of this thread if I knew how, as it has caused some confusion.
________________________________________________

This is a more thorough latitude test using some of the recently posted David Stump color charts.

Based on his notes I graded the base stop of f16 and then attempted to match that grade for f4, f5.6, f8 & f11 (for 4 stops of overexposure) and then f19nd.3, f19nd.6, f19nd.9 & fnd1.2 (for 4 1/2 stops of underexposure) - note that the f22 series were not posted at CML so I had to jump a 1/2 stop to the f19 series. All images were graded on discreet Inferno v5.54 using the original Color Corrector (not the Color Warper) - primary color grades only, not secondaries (or color grades through keys). All images processed at 1920x1080 from a scale/crop of original 4.5K.

[boothba]

PART 2:

This is similar to a CML comparison test featuring the Viper, the F900, Fuji F400 & 5218 film stock. Here:
http://www.cinematography.net/Pages%20GB/Film-Digi-comparison.htm

Consider this a 'quick 'n dirty' layman's analysis until Mr. Stump's findings are posted. Nonetheless, based on these images I would conclude:

1.) Red's ability to handle underexposure is exceedingly good - to my eye notably better than the Viper, F900, F400 & 5218 tests above. 2 1/2 stops under grades nearly identical to base stop, with very little digital noise. Likewise 3 stops to 3 1/2 stops under should be passable for a 2K or HD finish. 4 1/2 stops under shows more pronounced noise and warmth, but with secondary color grading & degrain filter it could certainly pass for a SD finish - even 720p finish. Ultimately the extremely low noise of RED is as important as the native latitude when digging deep into shadows - much better than 90+++% of film scans that I've dealt with.

2.) Red's handling of overexposure is less impressive on these images - surprising given what we've seen on the Porsche and milk shots. Mild clipping occurs at 2 stops over - but nothing major. Above that the clipping increases and the green channel and less saturated colors tend to go rather quickly. Also curious is that the extreme highlights seem to get overstepped by the near-highlights during overexposure. The good news is that Mr. Stump has indicated his belief that there may be an additional stop of latitude in the highlights by tweaking the existing color matrix.

3.) Based on these latitude tests (and until things change (count on it)) I would probably choose to underexpose a half to full stop when faced with tricky shooting situations - like a back lit actor against the sky with sun in frame. I think the sky would blowout long before the actor's face would get irreparably crushed.

4.) I've heard a few complaints that the images (and specifically the blacks) are milky - but I would argue that this is vastly preferable to having an aggressive matrix that crushes the blacks in camera. There is an enormous amount of retrievable data in the toe of the film (er,... digital image...) that the colorist can sculpt to his pleasing. Crushing the blacks will likely be the easiest damn thing to do in RedCine so no-one should complain. In digital cinema - FLAT is GOOD! Don't believe me? Check out a dMin/dMax 10 bit log scan for feature film DI and fx work. Or shoot a nice piece of 50ASA film and do a sensitometry test - or toss it up on a telecine bay at default settings and see what you get. Even telecine one-light transfers tend to add significant contrast so most folks are not really used to looking at pure film.

5.) Last point: regarding the color tearing at the right of frame in the darker exposures - we all know that this was a synch issue caused by a pinched cable - riiigghhhtttt???? - so no need to make Jim or Greame take the time to explain it again - riiiighhhhttt?????....

Well that's my two cents. I'm really impressed by these latitude tests - and if David is correct about possibly improving the highlights, RED should be ready to shoot black bears in the arctic - or deep-cave spelunking, incandescent albinos - or whatever.... Finally, If anyone would like a much larger version of the attached comparison chart (or the individual HD stills) let me know - I had do down-res for hvxuser specs.

[Jannard]

boothba... well said. We also believe we can improve the "over". And we also agree, along with other experts, that this camera can safely be set at ISO 320 to protect the top end without consequence. The lack of noise in the sensor helps in so many ways. We also agree completely that flat is good. It is easy to "snap up" the image from flat. Impossible to go the other way.

Jim

[Graeme_Nattress]

Remember there's no black offset calibration, hence the milkyness. Five seconds in a grade will fix that, and you won't see that anyway on camera raw or RGB images anyway.

We're still working on tweaking out the sensor. Also, remember dynamic range is the inverse of signal to noise ratio. With a digital system, nothing stops you clipping off the top, but you can just keep going down until you hit the noise floor.

Graeme

[_MR_]

This sensor though is more like a 5K sensor though is it not? What happens to the light sensitivity when you look at the production 4.5K chips? Surely being less tightly packed with light receptive electronic thingies the light sensitivity will increase?

If it does increase how might that effect the latitude?

[Graeme_Nattress]

Sensor is 4.9k. SHooting a 4.5k or 4k window from that doesn't change the size of the pixels.

Graeme

[Brook Willard]

Is the extra .4K visible in the EVF as overscan for the operator's sake?

[_MR_]

My bad - thought the test bed model was the only one with the sensor of 4.9K.

Got my memory wires a bit crossed.

[boothba]

Is the extra .4K visible in the EVF as overscan for the operator's sake?

This is a correspondence between Jim & Florian Stadler on the CML list:

"I want to see what's going on outside the frame lines.
I want to be able to spot that cable that's not hidden well enough...."

Florian Stadler, D.P., L.A.

You have that ability with RED. If you are shooting 4k, you WILL see 4.5k outside the recording area.

Jim

Jim Jannard
www.red.com

[taubkin]

Hey Boothba, thanks for the post.

It's cool that we can now skip technical numbers (like 10-11 stops of latitude) and get to the point of actual useable latitude. 8 and 1/2
stops of actual latitude is awesome, and something I can believe in. As it stands, it seems that the latitude is on par with film, and won't dissapoint anybody who is tired of video cameras clipping.

The thing that you can underexpose almost 3 stops without sacrifice (especially for SD or maybe even HDTV) is great! For music video production, that ends up on the small screen, has limited budgets, and a very high quality demand, I can't think of a better feature.

Good work with the tests, and thanks for sharing!

[Jannard]

David Stump determined that the Mysterium sensor is ISO 200, but he and many other experts will likely recommend shooting at ISO 320-400. That gives a ton of room for the highlights to survive and plenty of elbow room for a beautiful clean image.

Jim

[filmmaker1977]

so.. it won't be a elbow pain.. eheh..

portuguese meaning:
http://www.google.pt/search?hs=adw&hl=pt-PT&client=firefox-a&rls=org.mozilla%3Aen-US%3Aofficial&q=define%3Aenvy&btnG=Pesquisar&meta=

[Y.Goh]

Jim, do you think we can make it to 11+ f-stops of latitude DR when the camera is done next year?

[Jannard]

There is a big difference between dynamic range (11+ stops) and exposure latitude. Somebody else step in and explain it. I'm tired.

Jim

[Emanuel]

So, take a break, you deserve it more than someone else. For today, the good news are enough!

[Y.Goh]

http://en.wikipedia.org/wiki/Stops_%28Dynamic_Range%29 and http://en.wikipedia.org/wiki/Exposure_latitude, there are other literature too. Am I missing certain concepts here?

[Darkline]

Please do.

I am a little confused. If exposure lattitude and Dynamic Range are different terms, but we still speak of them in 'stops' lets talk about other cameras in terms of exposure lattitude for comparisons.

So far I've only heard about other cameras and film lattitude being described in terms of DR. For example, the HVX i think has 7 stops from what I've read, film 11 stops or more, I heard someone say they pushed a varicam to 10-11 but I find it hard to believe it's anywhere close to film's lattitude.

I dont mind which term is used, but it would be nice to have a reference for comparison. If Red has 8 1/2 stops of Exposure lattitude, what does a HVX have, film, and the current crop of high end HD cams?

[Y.Goh]

Darkline, I agree. I could be wrong here, but isn't DR is Signal to Noise Ratio, it means how much you can keep exposing light from noise floor 'till you get saturated/maximum signal. Exposure Latitude is how many stops of exposure from dark 'till you get highlight blows out. So what is the difference? Is that DR by its definition?

[Stephen Williams]

There is a big difference between dynamic range (11+ stops) and exposure latitude. Somebody else step in and explain it. I'm tired.

Jim

Hi Jim,

I will try!

Latitude is the amount you can over or underexpose and still get a good result.

A scene with a 7 stops of dynamic range, could in theory be over or underexposed by 2 stops and stay within the range of a sensor that had 11+ stops of dynamic range.

On a bright sunny day there may be in excess of 11 stops of dynamic range, so exposure accuracy can be very important.

Stephen

[grAde]

Boothba,
Your gradings of the original test chart frames were interesting and provided
an initial insight into the current pictures out of the Mysterium.
Id really like to see the same tests done with ramps as I find Grey
steps and Color blocks very hard to find things like Color Banding. Its
also a great way to see what happens to the values just beneath those clip points.

It would be great to see some:
Grey scale 0-100 ramps
FINITE white steps from (in 8 bit termilogy) from 220 - 255
Finite Black Steps from 0 - 25
Full 75%RED-75%GREEN Ramps
FULL 75%BLUE -75%YELLOW Ramps.

grAdo

[boothba]

If Red has 8 1/2 stops of Exposure lattitude, what does a HVX have, film, and the current crop of high end HD cams?

Just for clarity I'm not saying that Red has 8 1/2 stops of Exposure latitude. 8 1/2 stops simply represented the range of images I downloaded from CML. If they had posted the F22 series I could have gone further. I should have been more clear on that.

Once again this was just a visual experiment and not a professional assessment (I'm not qualified for that). Similar camera examples can be found here for comparison:
http://www.cinematography.net/Pages%...comparison.htm

BTW: HVX would not even come close - very noisey in the blacks.

Props to Red and David for letting us have a peek at these - very unusual for a camera under development. My guess is that we'll see more latitude tests shot in a few of months as things get closer.

[Y.Goh]

boothba, I read a post from another forum that might be an interest to the reader of this thread, and kind of agreed with you comment too:
'....I am a bit surprised that after a gain of 3 F-Stops I get an entirely white image.

I have a frame from DALSA that still shows image detail after a gain of 10 F-Stops and the SI frames I have still show image detail after 4 F-Stops of Gain.

What is striking is the absence of noise in the RED images - I wonder
if this has to do something with the lack of dynamic range.....'

[Jarred Land]

Dont judge the image range you see via thumnails on the Dalsa vs. the F950 latitude test on CML.. those thumbnails arn't an accurate representation of the comparison, the Dalsa ones have been pushed to reflect correct exposure, as soon as you download a frame you will understand.

[JDHolloway]

Agreed, those tests are interesting but not certainly not difinitive. It would be great to see a professional independant comparision against Dal/Vip/900 /D20/Kodak/Fuj etc.

While Red seems to have closed the gap (and passed it in many ways) with 35 mm, it still responds like video, abet cleaner.

Film...expose for shadows, print for highlights.
Video...expose for highlights, print for shadows (sort of)

Although I'm impressed by how much Gain can be introduced while still maintaining sharp edges, I find the colour accuracy even more impressive.

[Graeme_Nattress]

Umm.... It's not video that you expose to the right, it's digital (in general) you expose to the right. Digital photographers grok this - they use their histogram to ensure they capture their highlights, and as they have a good SNR, you know the shadows will be ok.

[JDHolloway]

Video...Digital...semantics surely.


I suppose "video" is a dirty word when your talking 4.5K! :beer:

[Graeme_Nattress]

Not entirely - with video, as you're shooting non-raw, you end up exposing for the middle :-)

Graeme

[donatello]

" It would be great to see a professional independant comparision against Dal/Vip/900 /D20/Kodak/Fuj etc. "

the test comparisions of the above camera's were done by a independent professional ...
here's a film out of f900, viper , kodak film , fuji film

http://www.cinematography.net/Pages%20GB/Film-Digi-comparison.htm

here's viper , cinealta, D20 , dalsa

http://www.cinematography.net/digital-latitude.htm

[JDHolloway]

Of course your quite right Graham, video (or jpeg for that matter?) RGB processing et al does do this doesn't it. I shoot quite a bit of RAW stills for kicks. But of course that's your specialty!

Silly me.

Donatello, I indicated that comparo in another thread "Whoops - I underexposed 4 stops!!! (http://www.dvxuser.com/V6/showthread.php?t=78124), but as RED was not there I hold it to be only partially usable as a test. Interesting, but only just. I feel the tests have to be done at the same time for controls sake.

[boothba]

boothba, I read a post from another forum that might be an interest to the reader of this thread, and kind of agreed with you comment too:
'....I am a bit surprised that after a gain of 3 F-Stops I get an entirely white image.
'

Hmmm... I would certainly NOT agree with that poster. I was able to extract a very decent image at three stops over as seen on the chart at the start of this thread. And that three stops over would become four over (or something) if the sensor was rated at ISO 320 - 400 as Jim has said.

Regarding Dalsa showing "image detail after a gain of 10 F-Stops" I've seen overexposed images hold onto the blacks loooonnnggg after the hightlights and mids are gone. Is that what is meant by "image detail"?

[Stephen Williams]

Hmmm... I would certainly NOT agree with that poster. I was able to extract a very decent image at three stops over as seen on the chart at the start of this thread. And that three stops over would become four over (or something) if the sensor was rated at ASA 320 - 400 as Jim has said.

Regarding Dalsa showing "image detail after a gain of 10 F-Stops" I've seen overexposed images hold onto the blacks loooonnnggg after the hightlights and mids are gone. Is that what is meant by "image detail"?

Hi,

The original poster is well respected on CML. I did not see anybody disputing what was said there.

Stephen

[boothba]

The original poster is well respected on CML. I did not see anybody disputing what was said there.

FWIW if the sensor is rated at 400 asa, the 3 stops over at 200 asa becomes 2 stops over at 400 asa.

Stephen

1.) Sorry, I don't know who actually said that (I am re-quoting Y.Goh above) but apparently the quote was "after a gain of 3 F-Stops I get an entirely white image." Well anyone can look at the original tiff and see that it is not entirely white. And with a color grade you can salvage all but the bright highlights. No disrespect to the respectable. I'm just using my eyes.

2.) Ya I butchered the ASA / ISO thing. For clarity here's Jim's original quote:

"David Stump determined that the Mysterium sensor is ISO 200, but he and many other experts will likely recommend shooting at ISO 320-400. That gives a ton of room for the highlights to survive and plenty of elbow room for a beautiful clean image."

:dankk2:

[Stephen Williams]

1.) Sorry, I don't know who actually said that (I am re-quoting Y.Goh above) but apparently the quote was "after a gain of 3 F-Stops I get an entirely white image." Well anyone can look at the original tiff and see that it is not entirely white. And with a color grade you can salvage all but the bright highlights. No disrespect to the respectable. I'm just using my eyes.

2.) Ya I butchered the ASA / ISO thing. For clarity here's Jim's original quote:

"David Stump determined that the Mysterium sensor is ISO 200, but he and many other experts will likely recommend shooting at ISO 320-400. That gives a ton of room for the highlights to survive and plenty of elbow room for a beautiful clean image."

:dankk2:

Hi,

You should hava a look at the CML thread, they talk about regrading, you may want to chime in.

Sorry, I messed up on the ASA, I deleted my error fairly quickly but not quick enough!

Stephen

[Stephen Williams]

David Stump determined that the Mysterium sensor is ISO 200, but he and many other experts will likely recommend shooting at ISO 320-400. That gives a ton of room for the highlights to survive and plenty of elbow room for a beautiful clean image.

Jim

Hi Jim,

From of what I have seen from David's tests I would go with 320asa which FWIW is the same I would expose film rated at 500 asa.

David mention's who calibrated his light meters, however he does not say what they were calibrated to! Different manufactures use from 11.5% to 18% which is another 1/2 stop in somebody's favour!

Stephen

[garth hagey]

good job, bootby. Elvino would be proud.





check your inbox, there's a pm for you.

[Stephen Williams]

boothba, I read a post from another forum that might be an interest to the reader of this thread, and kind of agreed with you comment too:
'....I am a bit surprised that after a gain of 3 F-Stops I get an entirely white image.
'

David Stump replied on CML as follows:-

"That's why I think it would be valuable to revisit the matrix that maps the sensor to RGB. I feel confident that there is plenty of latitude in there just waiting to be dug out with a little bit of curve noodling"

Stephen

[Jannard]

Greame has made an important discovery today that is yielding much improved results. We will do a re-shoot soon. Maybe we can talk David into coming down again. If not, we know how to light it this time. This is one of the things we opened ourselves up to by showing the work in progress. I'm good with it as long as people don't lock down what we are showing as final product.

Jim

[Stephen Williams]

Greame has made an important discovery today that is yielding much improved results. We will do a re-shoot soon. Maybe we can talk David into coming down again. If not, we know how to light it this time. This is one of the things we opened ourselves up to by showing the work in progress. I'm good with it as long as people don't lock down what we are showing as final product.

Jim

Jim,

Great News!
I hope you have found the comments from CML useful, possibly also saving you time.

Best wishes,

Stephen

[Graeme_Nattress]

This was an areas we were actively working in anyway, but Dave's scientific testing helps, as does the comments on CML (and here).

[Emanuel]

All productive inputs are welcome instead the counterproductive ones when they occur.

[Mr. Blonde]

If we only get 8 1/2 stops out of Red, that will be a dissapointment. :crybaby:

Let's get it back up there into the 11s Red team!

[Jarred Land]

dont worry.. ive seen the sensor record at least 11 stops (and then ran out of ND's) last week.

As David has said, he could have gone a couple stops both ways in his tests.

[Mr. Blonde]

That's what I like to hear. :)

[Mr. Blonde]

I'm also really hoping for an onboard, or atleast a Redraid camera mounted solution to 2K 120fps recording.

[Stephen Williams]

dont worry.. ive seen the sensor record at least 11 stops (and then ran out of ND's) last week.

As David has said, he could have gone a couple stops both ways in his tests.

Hi Jarred,

Any chance on posting the new frames?

Many thanks,

Stephen

[Jarred Land]

We are shooting new frames.. things have improved since then (2 weeks ago) so the frames posted next should be able to show a better range. It would be good to be able to have a single frame latitude test rather than shooting 30 frames.

[Stephen Williams]

We are shooting new frames.. things have improved since then (2 weeks ago) so the frames posted next should be able to show a better range. It would be good to be able to have a single frame latitude test rather than shooting 30 frames.

Hi Jarred,

With ND 0.15 gell you could build a backlit grey scale with 0.5 stop intervals. Done that before now!

Look forward to seeing newer tests,

Stephen

[Anders Holck]

http://www.stouffer.net/TransPage.htm seems to be the most scientific way of doing range tests if all you want to know is the absolute maximum range.
T4110CC offers 13 stops in 1/3 stop increments.

[Graeme_Nattress]

Anders, I'd agree that that's the best way to do a nice, single shot dynamic range test, as then you can see where the white's blend together and the blacks become indistinguisable, and then just read off the answer in stops. Easy. That's the technique that DPReview uses - it's accurate and repeatable, and not open to much subjectivity.

Graeme

[Jarred Land]

absolutely.. I think we should try and make that test the standard.. its much more non-subjective as Graeme says, and is much more fitting for digital sensors.

[Graeme_Nattress]

No complaints from me Jarred :-)

Graeme

[Anders Holck]

Have you guys tried to contact Dpreviews Phil Askey?
He seems to have some great practical experience with testing procedures both with regards to noise, sharpness, dynamic range etc.
Especially if you get a shootout done in the new year, it would be very good to be able to show noise curves with both black and white patch to compare againt sharpness patches and responce curves etc.

[ZaneIsNumber1]

Here's a six camera comparison posted over at cinematography.net. Includes the Red camera, although we know those results are already out of date.

http://www.cinematography.net/hdcamtests/HDcurves.htm

[Graeme_Nattress]

And I know that the RAW to RGB conversion was not correct for those images too.

Graeme

[ts1]

And I know that the RAW to RGB conversion was not correct for those images too.

Graeme

Graeme,

Have your revised the debayer algorithm as well? While we are on the topic of debayering, how sensitive is your algorithm to chromatic aberration? It doesn't take much TCA for the RGB correlation model to break down.

Regards,
TS

[Graeme_Nattress]

There's multiple algorithms on the go at the moment.

Graeme

[ts1]

Perhaps at a later date you could support TCA correction in Redcine. This will allow the less wealthy users to use cheap glass without sacrificing debayer performance.