is UMPro officially better in low light? See FPN samples.

Considering that both cameras have the same sensor, I was sad when I noticed, that my UM Pro had a slightly worse FPN, than the UM46. I expected a better SNR on the new model.

So in order to guess if my unit is just a bad luck, I want to ask a straightforward question:

Did anyone from Blackmagic stated anywhere, on a forum, or in an interview, that the UM Pro should be better in low light, than the older UM46 ?

Sorry if the topic is not very useful, I am just worried. Thanks.

I have several copies of both and I've not noticed a particular difference.

There are of course easily seen differences between individual copies of the cameras. Each sensor is slightly different, but as I understand the sensors are the same, as is the fundamental hardware.

But each sensor itself will be a little different. One of my earlier Ursa Mini's needs a tint correction of something like -39 to make it match the other Ursas.

I also see the same thing in other cameras. I have four Alexas on the show I'm shooting now and none are the same, they each have a slightly different LUT to normalise them for my dailies.

JB

What Emilian is describing isn’t a standard color balance difference. I’ve compared uncompressed raw shots from two different Ursa Mini Pro and both have a blue tint in underexposed blacks that’s not there in raw shots from the original 4.6K, and it’s not possible to just dial it out by adjusting temp and tint in the raw tab of Resolve. There was definitely something different about the noise in those Pro units, even after properly running the shading calibration multiple times. Maybe those two UMP sensors are a fluke. All I can say for certain is that they didn’t perform the same as my older 4.6K.

You can see details about my worries in the other topic here:

viewtopic.php?f=2&t=68094&p=381278#p381278

but I want to keep this topic clean. Its purpose is just the answer of a single question.

Thanks.

I take it that the Pro version is the fixed version of the 4.6k, meaning that the fixed itself probably make a little difference. It's been only a year or less since the original hit the streets till the second did, wasn't it? So I don't expect the underlying technology to have changed much, but the front of the sensor might have, and with calibrations for better overall performance, it might be receiving less light and/or be slightly worse or more gsinedst lie light (does it reach down lower, or is there more noise at the same picture response?).

FPN comparison at 1600 ISO, very underexposed.

UM46 EF:

https://drive.google.com/file/d/10Llfma ... 7TqIr/view


UM46 PRO:

https://drive.google.com/file/d/1uDhKMn ... AF4Yl/view

My UMP PRO is better in low light than my 4.6, (which is still not bad) My wife's ump pro is worse than either of mine.

WOW for each his/her own UM Pro - you are one happy family

So I guess, with the new model its a game of chance, as it was with the old one...

Still I just need an official answer from a BM rep:

Is the UM Pro supposed to be better, or not?

Based the available empirical evidence I think we can make a reasonable conclusions about the answer:
The tolerance allowed in BMD's QC standards for the 4.6K sensor is wide enough that the differences between sensors can be seen.

Manufacturing tolerances are a reality for any device. BMD gets to set them however they want, they can change them anytime they like, and they don't have to tell us a thing about it.

Considering that with a bit of careful exposure and some skill in post you can get images out of the 4.6K that rival an Arri Amira for 1/10 of the price, BMD has got to be cutting some costs somewhere relative to Arri. Using hardware that more stringent standards would have otherwise rejected, thus increasing yield, is one way to cut costs.

That's the trade off with these cameras. I seek the best image I can get along with the easiest post workflow, but I don't have $50K to throw at Arri. And there really just aren't any other cameras out there that can shoot 4K 12bit 444 ProRes files. All the other options drop down to 10bit 422 (in various codecs) at 4K or force you into less convenient raw workflows. Red's .r3d files have a relatively easy workflow (though you need a pretty powerful computer), but building out one of their cameras with accessories to get the same features as an Ursa Mini Pro easily takes you north of $20K. And Red's profile process for getting FPN out of their sensor is a doozy. The tutorial video on how to do that that Shane's Inner Circle offered for their members is over an hour long. Not something you want to be doing before every shoot.

I'm totally completely with you that it's a frustrating situation. It sucks that I can't just order another 4.6K and be completely confident that it will perform identically to the one I already own. But I'm still grateful that BMD is in the game and doing what they can to deliver amazing images relative to the price point.

Thanks for the assurance Jamie, I can agree 100% on all your points!

Just wanted to check if anyone from BM ever mentioned in an interview or whatever, that the UM Pro should be better. If they did, I would know, that I can hope for a better unit down the line.

Just stumbled on this review and WOW, these guys seem to have a UM Pro, without ANY FPN - is that even possible? FPN on my unit is pretty visible, so is it a game of chance again?

https://goo.gl/p89XYU

Just stumbled on this review and WOW, these guys seem to have a UM Pro, without ANY FPN - is that even possible? FPN on my unit is pretty visible, so is it a game of chance again?

https://goo.gl/p89XYU

Wow. That doesn't match my experience with the camera. 5 stops they show being able to raise the exposure without FPN. Perhaps they made a mistake somewhere? By their own admission, they only had the camera for a day and had zero previous experience with it.

Yes, seems impossible. The only scenario I can see such an achievement, is if one shoots at Prores 444, ISO 200 in "video mode". That means the blacks will be mostly crushed and clean, but still, 5 stops...

with the non pro 4.6k I have been able to lift 8 stops under on skin and clean up the FPN with neat video to a useable place and 5 stops over. I have been able to lift about 4.5 stops without any noise reduction or repair. 5.5 over I might be about to recover skin but it would be tight.

I cannot release those frames but I will try and do another test so that I can share.

Hi Ryan, can we test this numerically in IRE values? I think that way we could compare the sensors more accurately.

So you say, that you can bring back skin, from 4 stops under. Without any FPN.
If we assume the skin tones are properly exposed at 60-70 IRE, then you should be able to bring back skins exposed at 15-17 IRE.

Then maybe your sensor gets FPN at 10-12 IRE.

If this is confirmed, I can also check my sensor at what IRE values, the FPN starts to be present on skin tones.

My camera is on rental through the weekend but I should be able to confirm after that.

So I did an exposure test, researching the noise on skin tones on a black background.

The subject is slowly moving within the frame, so the dreaded FPN vertical lines of madness are easier to spot.

I did a proper exposed reference shot with skin tones around 60-70 IRE at F2.8.
Then I did under exposed shots by 1 stop increments at F4.0, F5.6, F8.0, F11, F16.

Here is the file for download:

https://drive.google.com/open?id=1bS6hh ... NbyIt2wKSB

Notes:

Ursa Mini Pro, firmware 4.7, black shading on, RAW 4.6K, ISO 800, no color correction, just processed in Davinci and converted to Prores 422.

0. Proper exposure - skin is great, background is great, no color shift.
1. Under exposed by 1 stop - skin is great, background noise is alright, FPN can be seen in the darker parts of the hair but it is very slight, no color shift.
2. Under exposed by 2 stop - skin is good, regular noise is usable, FPN can be seen in the darker parts of the hair, very slight color shift.
3. Under exposed by 3 stop - skin is good, noise and FPN are bad in darker areas, color shift is now visible.
4. Under exposed by 4 stop - skin is noisy, noise and FPN can be seen at dark and midtone areas, color shift is bad.
5. Under exposed by 5 stop - skin is bad, FPN is bad, everything is bad.

Conclusion:

My Ursa Mini Pro can be pushed by 2 stops for a usable image, but only by 1 stop for a clean image.
If I push it for 3 or more stops, the FPN and color shift become visible.
Overall, the FPN starts to show up at values below IRE 25 and gets bad below IRE 20.

If someone from Blackmagic could check on these results, it will be great to get a feedback.

So I did an exposure test, researching the noise on skin tones on a black background.

The subject is slowly moving within the frame, so the dreaded FPN vertical lines of madness are easier to spot.

I did a proper exposed reference shot with skin tones around 60-70 IRE at F2.8.

Ursa Mini Pro, firmware 4.7, black shading on, RAW 4.6K, ISO 800, no color correction, just processed in Davinci and converted to Prores 422.

So this is just FILM footage converted to ProRes ?

I'm on a slow connection here so I can't download it.

When I look at you image as a preview in Google, nothing is black.

Look at the BLACK of the surround from the web window that presents the file when you click on the link. Let's call that TRUE black. Let's call that REFERENCE black.

Nothing I look at in you image contains TRUE black, including you black background or the shadows from your subject being cast on the black background.

In my view, the BLACK of you fabric should be sitting JUST above black on the waveform.

The shadow being cast by the head should be at TRUE black,

It's not meaningful to judge for noise or FPN when looking at LOG images on a computer screen like this.

Are you saying skin tones are at this tone in LOG ? In LOG that should actually be somewhere in the low 30's not 6-70 which is actually quite over exposed (in log)

JB

Here's what I'm seeing. This is only a screen shot from the google chrome browser window but you get the idea.

Then I've brought down the blacks and the mids in FCP, plus lifted the whites a bit and added some chroma.

This is what you should be then looking at your results with. Unless you plan to grade your footage with the blacks so lifted ?

JB

Ref1.png (230.68 KiB) Viewed 4452 times

Ref2.png (236.25 KiB) Viewed 4452 times

John, thanks for the observation.

Yes this is straight out of the camera.

Please consider that the background looks grey, because it is so close to the subject, it is being lit by the light as well. So the blackest blacks should be at the shadow behind his head.

The reason to keep the blacks lifted - it is very easy to see the difference between the FPN at +1, +2, +3 stops, etc. And I wanted to see at what IRE levels, the FPN starts to show up.

Sorry for my ignorance, but I am not aware about exposing in LOG, why the skin should be at IRE 30?
I exposed the skin at IRE 60-70 in camera, using the Video Assist monitor waveform.

The skin looks perfect on my monitor.
For reference, I switch to 8-bit mode and these are the 8-bit RGB values:

SKIN: 135, 115, 95
Background: 30, 30, 30
Shadow behind head: 8, 8, 8

The 8,8,8 is close to true black. So there is a leeway for grading.

Will it help to upload the RAW files?

Scopes from Davinci.

The waveform is showing values from 0 to 1023, and the brightest skin tones are around 640-800. Which is in the ballpark of 65-75 IRE.

P.S. does anyone know, how the set the waveform to show IRE values from 0 to 100 ?

Sorry for my ignorance, but I am not aware about exposing in LOG, why the skin should be at IRE 30?
I exposed the skin at IRE 60-70 in camera and then the waveform i showing just that.

I'm never one to say where skin tone's should be exposed because that's the job of drama. But in a TV studio somewhere in the 1970's some engineers decided that 75% was a good IRE for skin tones on daytime talk television.

Blackmagic's MID grey point in their BMD Film log is 37.2 IRE. So when looking at a bmd LOG image your skin tone should be somewhere around the high 30's at most if you want to replicate this thinking.

If you've got skin tones in the 60-70 range you're putting skin more than 2 stops(ish) above mid grey.

The bigger point is really, look at your image.

Is that really how you want your skin tone to look ? Is that how you want your image to look ? Don't you think my FCP grade on your screenshot looks better than what you presented as a normal image ?

Conservatively I pulled your exposure down by two stops.

JB

If mid grey sits at 37.5 (caucasian) skin tones should sit somewhere in the 45-55 range unless your subject is deeply tanned. Where you put them artistically is another matter. Either way pushing couple stops on a well exposed image shouldn't present a problem for an image that overexposed to begin with. There's either a problem with your workflow or your camera. Post a couple DNGs, I'd suspect the former.

If mid grey sits at 37.5 (caucasian) skin tones should sit somewhere in the 45-55 range unless your subject is deeply tanned. Where you put them artistically is another matter.

Howie I think this is true only if you want to match this old fashioned TV studio thinking of skin tones needing to be at 75% when looking in a REC 709 like space.

I've always found caucasian skin tones look great on the Ursa Mini 4.6K when they're sitting closer to mid grey (greens and a teeny edge of pink in FC)

I think Emilan you need to understand that in LOG the way you map exposure is different to a more VIDEO style space of 709. 75% in REC 709 isn't the same as 75% in BMD LOG by a lot.

And it goes back to my original point.

When you have a normalised image where blacks are actually black (demonstrated on a waveform !) and you're looking at it on a TV monitor (YUV) rather than a Computer monitor (RGB) things begin to look very different.

Your normal exposure is in fact overexposed by two stops maybe, so you're then trying to lift it again in post and then viewing the image in a way that will be guaranteed to show faults that get buried in the blacks of a normalised image.

Perhaps Emilian you can use the FC in the camera itself as it's not affected by LUT's etc and gives you a closer idea of your true exposure instead of using a waveform in an external monitor.

JB

Put skin tones wherever you like, I don't think this is an artistic discussion. We could also talk about the color but again this is not an artistic discussion. If you expose mid grey at 37.5 skin tones will sit higher I don't know how to explain it any more clearly.

If you watch the video, I'm guessing, you haven't, the OP brackets exposure stopping down and pushing it back up accordingly. By the third stop of push/pull the skin tones are sitting between 35-40 IRE before correction and the image has pretty much fallen apart, though it is still very well exposed. Something is wrong with the workflow or the camera. Unless of course this is expected behavior, in which case everything is fine.

I primarily shoot raw log with skin varying between green and pink false colour depending upon dynamic range in the scene and the light falling on the skin. Pink is a stop above green (middle grey).


Sent from my iPhone using Tapatalk

Thanks for all the info, I never though exposing the skin like that. I am used to ETTR.

Anyway, I still think that my camera will produce bad FPN, if the skin is exposed at 40.

Here are the DNGs from the test:

https://drive.google.com/file/d/13YYRcM ... Dfg6nlmqk/

These are 5 shots, exposed as described: EV0 (skin 60-70 reference), EV-1, EV-2, EV-3, EV-4, EV-5

So should I expect to get proper skin tones out of the "EV-2" or the "EV-3" shot?

* Only open in Davinci, I guess it was RAW 4:1.

Oh and Happy New Year! Lets wish for less FPN during the next year

Happy new year Emilian.

The skin tones can be where you like. But what you presented as a normal exposure isn’t what I would call normal.

You’re expecting no FPN but all your blacks are starting wayyyy above black and your skin tone is also very high. Your skin tone being high along with your blacks being high says you’re two stops over NORMAL to me.

You say you’re seeing FPN about two stops before everyone else does so......maybe the way you expose is what’s causing your FPN to show up.

If you interrogate a LOG image with lifted blacks FPN will be there.

My question again to you is, what if the grade is more normalised ?

The thing is Blackmagic Is amazingly resilient in that you can routinely overexpose by two stops like you are and get a normal image without consequence. You can’t really as easily do this with many other cameras.

JB

Are you saying skin tones are at this tone in LOG ? In LOG that should actually be somewhere in the low 30's not 6-70 which is actually quite over exposed (in log)

JB

John, just to confirm, are you really exposing skin (in log) below middle-grey?

I realize this exposure would bring the normalized skin tone levels to maybe 45-55 ire, which should be fine, at least on a BMD camera, but it still seems to run counter to conventional wisdom. Can you elaborate?

Are you saying skin tones are at this tone in LOG ? In LOG that should actually be somewhere in the low 30's not 6-70 which is actually quite over exposed (in log)

JB

John, just to confirm, are you really exposing skin (in log) below middle-grey?

I realize this exposure would bring the normalized skin tone levels to maybe 45-55 ire, which should be fine, at least on a BMD camera, but it still seems to run counter to conventional wisdom. Can you elaborate?

As I alluded to earlier in this thread, this thinking about skin tone being at 70% has never made sense to me simply because it's based on the idea of TV studio scenarios...aka the news.

Sometimes skin tone is in shadow. Sometimes it's in the blinding glare of sunlight. Sometimes I've got someone that's very dark and very pale in the same frame.

I think of exposure as a choice, not something that's right or wrong or correct.

I use false colour as a guide to see where everything is sitting. Skin tone is one (important)part of the exposure choice. It's also why I like Zebras as it can tell me what's absolutely clipping when set to 100% while shooting mid-take.

I also shoot a lot of ProRes Alexa. With Alexa, I have noticed it's actually less linear or HARDER to recover near overexposed detail compared to the Ursa.

So it's well known that Alexa almost can't clip and yet if you put important exposure information in the higher range on Alexa (like skin tone) then you can't just as nicely recover an exposure of skin at say 90% back down to 50%. It just seems kind of thin. It's ETTR but it just doesn't work in the extremes of exposure.

So, in my experience, shooting ProRes 444 on Alexa even though it's technically not clipped the curve applied is such that it isn't really possible to magically pull everything back down without it looking (grading) inferior to if you exposed it that way in the first place.

I call or think of this as the linearity of the camera. How evenly does the information in the shot hold as you scale up and down the exposure scale.

So when I do bracketed exposure tests, I try to set a grade I like then I see how malleable the shot can be as I work up and down through the exposure brackets. If the colour and detail doesn't hold as the exposure changes then I consider it to not be tracking or not linear.

We know this already as discussed in this thread for example, that colour information changes with exposure . More saturation as you under expose, less colour information as it goes towards clipping.

So I tend to pull exposure down generally a little on skin tones with Alexa to protect them and I'm usually using that as my main camera that I'm then shooting Ursa alongside, so I tend to match the Ursa exposure to Alexa.

jb

John, thanks for you advises. Maybe I should not expect for a black background, to look like a grey one, even when lit. I mean, it should be black, so the way you graded it looks just fine.

The thing that bothers me, is the difference between the test results. It seems you have much more experience with different RAW formats, than I do. Could you please open the DNGs and check which one is most properly exposed by your view? Maybe the EV -2?

Are you saying skin tones are at this tone in LOG ? In LOG that should actually be somewhere in the low 30's not 6-70 which is actually quite over exposed (in log)

JB

John, just to confirm, are you really exposing skin (in log) below middle-grey?

I realize this exposure would bring the normalized skin tone levels to maybe 45-55 ire, which should be fine, at least on a BMD camera, but it still seems to run counter to conventional wisdom. Can you elaborate?

Exposing anything at 30 IRE in log will still be at around 30 IRE after it's "normalized" using either the V3 LUT or a CST. The pivot, for lack of a better term sits right at about 30-32 IRE so anything above gets pushed up and anything below gets pushed down.

Emilian:

Here are a couple grabs from your footage. When you get to the point that the skin drops below 50 IRE in log your blacks are pretty much toast. The FPN becomes apparent even after simply applying the V3 LUT. Push even a single stop and it's obvious. I would guess this would be irregular behavior. I'd forward your DNGs to BM and have them take a look.

Good Luck


LOG

As Shot Log.jpg

BMDFilmtoRec709V3

BMDFilmtoRec709V3.jpg

BMDFilmtoRec709V3 Pushed 1 stop

BMD4.6toRec709+1.jpg