Page 1 of 1

low light performance compared

PostPosted: Thu Dec 21, 2017 5:59 pm
by Benjamin de Menil
I work a lot in low light and have been wondering about getting a B camera for it. So far the BMMCC has been my go to for low light. Shooting in RAW, I can boost quite a bit. The max iso equivalent is several stops higher than the BMMCC's native 800. I'm curious how others would quantify it. In the world of DSLR and mirrorless people talk about max usable iso, and don't mention native ISO much. A lot of cameras begin to hit their limits at iso 3600. I wonder if they're any better than the micro cinema...

For instance, how might the micro cinema compare to sony's A7r ii? I've seen some posts saying that the A7r ii has the equivalent of two native ISOs: 100 and 640. The micro's native is higher and it can record with less compression. Does this mean the micro is better in low light, or is the micro noisier at it's native than the sony? Or does the sony use a process for boosting iso that is better than boosting gain of the micro in post?

Re: low light performance compared

PostPosted: Thu Dec 21, 2017 7:10 pm
by Robert Niessner
Most so called low-light cameras use in-camera noise-reduction. All Sony cameras do that and all Canon cameras do that.
Blackmagic cameras have NO in-camera noise-reduction, but with in-post noise-reduction you can push the limits a lot.

Re: low light performance compared

PostPosted: Thu Dec 21, 2017 7:54 pm
by Dmitry Shijan
Image
Image

I done some extreme deep into the shadows tests with BMMCC. Timelne set to BMDfilm Log. Boosting Expose and than lowering Gain produces great result. It keeps highlights unchanged and recovers information from deepest shadows. This s a 500W halogen lamp in almost dark room. ND64 filter on the camera. You can do the same Expose boost with ProRes by adding Gain node in linear space. It is sad but FPN is the main limit in this camera and i still try to find a better way to remove it with manually created FPN sample patterns.
BMMCC also don't have additional pixels for 2x, 4x or larger downsampling to smooth noise artifacts.
Image
Image
Image
Image
Image
Image
Image
Image
Image

One more Expose boost test. This is BMPCC RAW sample from online camera review. Timeline set to BMDfilm Log. Expose 6 created by setting ISO to 1600 + Expose 5. Also applied my own Color Checker correction node from BMMCC.
Image
Image
Image
Image

Re: low light performance compared

PostPosted: Thu Dec 21, 2017 9:04 pm
by Benjamin de Menil
Am i right that with the BMMCC's 4:2:2 sampling, color becomes compromised with extreme shadow boosting like what Dmitry is doing? In the first example there isn't much color so it's not noticed...

Isn't boosting exposure and lowering gain similar to boosting shadows using the node wheel or curve?

In resolve does each integer increment in exposure correspond to 1 stop?

Re: low light performance compared

PostPosted: Thu Dec 21, 2017 9:13 pm
by Benjamin de Menil
Robert Niessner wrote:Most so called low-light cameras use in-camera noise-reduction. All Sony cameras do that and all Canon cameras do that.
Blackmagic cameras have NO in-camera noise-reduction, but with in-post noise-reduction you can push the limits a lot.


Cameras with a reputation for good low light generally do have high native ISO. A7S is iso 1600 for instance. If the camera can add gain to raw image before sampling to 8bit 420, it theoretically has a 1 stop advantage over bmmcc, more if micro has noisier shadows which i suspect it does.

Re: low light performance compared

PostPosted: Thu Dec 21, 2017 9:33 pm
by John Brawley
Benjamin de Menil wrote:Am i right that with the BMMCC's 4:2:2 sampling,


Sampling is a function of encoded video.

So ProRes 422 is 4:2:2 encoded, however, if you shoot RAW then you are not limited to 4:2:2 encoding.

You may be conflating 4:2:2 encoding with the RGB ratio of pixel sampling within a BAYER sensor, which funnily enough is around the same ratio of green to red and blue pixels.

This is a totally different proposition though and it's not as simple to extrapolate this kind of information from the numbers.

Dmitry has done some great work exploring the boundaries of what can and can't be done in low light.

I know that many have found that ISO is a misleading number. When you have a RAW shooting camera you can judiciously apply noise reduction and careful grading tailored to each shot so that you can push in post the apparent ISO and still have a better looking image than a camera that has an ISO setting that goes to 11.

I know that many have found that their Sony's et al that can shoot in the dark don't necessarily make a nice picture in the dark.

JB

Re: low light performance compared

PostPosted: Thu Dec 21, 2017 11:11 pm
by Greg Lee
Pardon my ignorance, but when you say "Boosting Expose", what are you referring to? I know of the "Gain" dial but I'm not familiar with an "Expose" control...

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 12:21 am
by Jamie LeJeune
John Brawley wrote:I know that many have found that ISO is a misleading number. When you have a RAW shooting camera you can judiciously apply noise reduction and careful grading tailored to each shot so that you can push in post the apparent ISO and still have a better looking image than a camera that has an ISO setting that goes to 11.

I know that many have found that their Sony's et al that can shoot in the dark don't necessarily make a nice picture in the dark.

JB

+1
Same experience here. A couple years ago I shot with a Canon C300, a Sony FS7 and the original BMCC in the same challenging low light on a documentary. Going into post I thought the BMCC would be the weakest image in the darker scenes, and was surprised to find that in nearly every case the BMCC yielded the best image, even though the Canon and Sony were recording a higher ISO.

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 2:28 am
by Benjamin de Menil
+1 on Dmitry's test being useful.

@John Brawley, I read somewhere about the RGB pixel sampling and you're right I did conflate the two. But the conclusion is the same, as you boost the shadows, the color seems to break up before the luma. Zooming in on Dmitry's photos you can see the color noise. I've found it challenging to get color looking good when BMMCC footage is under exposed. On the BMPC it's even more of a challenge. And I haven't found that RAW is any better the Prores HQ on the BMPC.

I agree there's a lot of spin around some of the high ISOs in the camera world. But it's not always spin. Some cameras are genuinely very sensitive.

Jamie, your experience with the FS7 and C300 is interesting. Those are two cameras I was considering. I'm amazed the BMCC beat them in low light.

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 2:40 am
by Benjamin de Menil
Greg Lee wrote:Pardon my ignorance, but when you say "Boosting Expose", what are you referring to? I know of the "Gain" dial but I'm not familiar with an "Expose" control...


In davinci resolve's raw controls there's an exposure setting that I think is about the same as boosting gain within a camera. I may be wrong, but I think that the gain control in Resolve boosts the brighter part of the image more than the shadows, but that exposure boosts everything. I'm not sure which corresponds most closely to in camera gain controls.

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 3:11 am
by John Brawley
Benjamin de Menil wrote:+1 on Dmitry's test being useful.

@John Brawley, I read somewhere about the RGB pixel sampling and you're right I did conflate the two. But the conclusion is the same, as you boost the shadows, the color seems to break up before the luma.


I wouldn't attribute it to the BAYER sampling, but if you're unhappy with the chroma noise then it doesn't really matter how it gets there.

Benjamin de Menil wrote:I agree there's a lot of spin around some of the high ISOs in the camera world. But it's not always spin. Some cameras are genuinely very sensitive.


I don't think it's spin. They work no doubt. The question really is, how do you like the picture that comes out ?


Benjamin de Menil wrote:Jamie, your experience with the FS7 and C300 is interesting. Those are two cameras I was considering. I'm amazed the BMCC beat them in low light.


Exactly the point I make.

JB

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 5:40 am
by Dmitry Shijan
Benjamin de Menil wrote:Am i right that with the BMMCC's 4:2:2 sampling, color becomes compromised with extreme shadow boosting like what Dmitry is doing? In the first example there isn't much color so it's not noticed...

Isn't boosting exposure and lowering gain similar to boosting shadows using the node wheel or curve?

In resolve does each integer increment in exposure correspond to 1 stop?


I do same tests with RAW and with ProRes422 and don't notice dramatic difference in shadow details boost. I don't use ProRes422 only because it baked with factory input color correction so it looks more saturated and greener than RAW. Same time RAW looks dull and pastel without manual color correction, but allow to manipulate with 100% original untouched data.

Boosting Expose for RAW is 100% same as boosting Gain for ProRes in Linear Gamma (see node chain in first post).
Lowering Gain must be done in LOG gamma. Each LOG gamma have different contrast response, BMDfilm and RED Log3G10 works best for this adjustment in my opinion.

To adjust Exposure in ProRes footage same as in RAW add two Color Space Transform nodes, and one more node in the middle.
Set first node input gamma Timeline Gamma in , Linear out
Set another node input gamma Linear in , Timeline Gamma out.
Adjust Gain in the middle node. This produce exact result as Exposure slider in RAW controls.


Boosting exposure and lowering gain is similar to boosting shadows but works in more natural and physically correct way.
Boosting shadows/highlights in Secondaries tab don't designed for so large and uniform adjustments.
Boosting shadows/highlights slider in RAW tab works same as Adobe Camera RAW or many other RAW processing apps. It adds destructive luma masks and ruins your footage source. So be aware of use it.

About 3 additional stops recovered from shadows is the real clean limit for BMMCC camera. Dig deeper and Fixed Pattern Noise became too visible over moving image.
Feel free to underexpose. But you are forced to use temporary noise reduction mixed with Chroma noise reduction, adding 0.5 XY shift to remove Cross Hatching and using OLPF+IR filter to eliminate wired moire/aliasing in RAW. Also use RAW+ColorChecker correction to get neutral starting point without shifted and desaturated colors. This will allow 99.5% pixel perfect quality from BMMCC camera without downsampling.

See this threads for more detailed info about workflow: https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=65149
https://forum.blackmagicdesign.com/viewtopic.php?f=2&t=66184

Image
Image
Image
Image
Image
Image
Image

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 9:42 am
by Dmitry Shijan
BTW here is a sample from Canon C200 RAW. As you can see without camera noise reduction and downscale to HD it is full of noise in shadows. RAW samples from DJI drone even more ugly when you start to boost shadows information.

Image
Image

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 2:51 pm
by Wayne Steven
Robert Niessner wrote:Most so called low-light cameras use in-camera noise-reduction. All Sony cameras do that and all Canon cameras do that.
Blackmagic cameras have NO in-camera noise-reduction, but with in-post noise-reduction you can push the limits a lot.


Robert, various sensors have in sensor pad circuit noise reduction. Maybe a preferable starting point. I think Sony might be one that does.

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 3:11 pm
by Benjamin de Menil
John Brawley wrote:
Benjamin de Menil wrote:+1 on Dmitry's test being useful.

@John Brawley, I read somewhere about the RGB pixel sampling and you're right I did conflate the two. But the conclusion is the same, as you boost the shadows, the color seems to break up before the luma.


I wouldn't attribute it to the BAYER sampling, but if you're unhappy with the chroma noise then it doesn't really matter how it gets there.

JB


My issue is the lack of color in the shadows when I boost them. It look to me like there was less color info captured in the shadows, and that comparatively there is more luma info. I have a harder time getting the color right on under exposed footage. I'm new to this so tell me if it's my eyes playing tricks on me.

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 3:18 pm
by Benjamin de Menil
Thanks for that info Dmitry. Looking forward to studying those workflow threads. I've been using lift, gain, and highlight in the raw controls. Didn't realize these were destructive edits.

That mosaic filter looks amazing too. Any downside to it?

Re: low light performance compared

PostPosted: Fri Dec 22, 2017 8:19 pm
by Dmitry Shijan
Benjamin de Menil wrote:
John Brawley wrote:
Benjamin de Menil wrote:My issue is the lack of color in the shadows when I boost them. It look to me like there was less color info captured in the shadows, and that comparatively there is more luma info. I have a harder time getting the color right on under exposed footage. I'm new to this so tell me if it's my eyes playing tricks on me.


after you apply "rise expose-lower gain" you get hdr-like low contrast and low saturation image. try to add contrast and you will see more saturation

Re: low light performance compared

PostPosted: Sat Dec 23, 2017 1:40 am
by Uli Plank
It's logical that colors disappear close to the lower and upper end of the contrast range.
After all, black is zero color, just like white.

Re: low light performance compared

PostPosted: Sat Dec 23, 2017 10:58 am
by Benjamin de Menil
Uli Plank wrote:It's logical that colors disappear close to the lower and upper end of the contrast range.
After all, black is zero color, just like white.


It's logical as long as shadows remain shadows. I'm boosting them into the middle of the range.

Re: low light performance compared

PostPosted: Sat Dec 23, 2017 12:38 pm
by Robert Niessner
Benjamin de Menil wrote:
Uli Plank wrote:It's logical that colors disappear close to the lower and upper end of the contrast range.
After all, black is zero color, just like white.


It's logical as long as shadows remain shadows. I'm boosting them into the middle of the range.


Boosting in post does not change where the shadow colors had been for the sensor during exposure.

Re: low light performance compared

PostPosted: Tue Dec 26, 2017 6:17 am
by Benjamin de Menil
Robert Niessner wrote:
Benjamin de Menil wrote:
Uli Plank wrote:It's logical that colors disappear close to the lower and upper end of the contrast range.
After all, black is zero color, just like white.


It's logical as long as shadows remain shadows. I'm boosting them into the middle of the range.


Boosting in post does not change where the shadow colors had been for the sensor during exposure.


Aren't the colors still physically there in low light, just a weaker signal? Color is interpreted somehow in debayering. It just looks to me like the luma info survives boosting better than color. Not sure why.

Re: low light performance compared

PostPosted: Tue Dec 26, 2017 7:29 am
by Uli Plank
Yes, colors exist, but the signal is about one stop weaker for red and blue than for green (with a Bayer pattern). Consequently, you'd need more amplification for correct colors and get more noise. Or you reduce color in favour of less noise.
There are similar problems close to saturation: one or two colors may clip earlier and you get a cast in the complement color. The only solution is reducing saturation close to clipping.Colorists have known this trick for years: copy the channel that's not clipped into the saturated ones (only for the highlights). You loose color, but gain detail.

Re: low light performance compared

PostPosted: Tue Dec 26, 2017 7:43 pm
by Benjamin de Menil
Uli Plank wrote:Colorists have known this trick for years: copy the channel that's not clipped into the saturated ones (only for the highlights). You loose color, but gain detail.


I've read about that and want to learn how to do it

Re: low light performance compared

PostPosted: Wed Dec 27, 2017 2:28 am
by Uli Plank
Well, it's easy: make a mask for the highest luma values ( soft edges, as always in grading) and use channel operations to copy the non-saturated channel into the other two. Done.
BTW, Red is doing just the same with their DRx function.

Regarding color in the shadows: audiences will normally find colored noise more objectionable than luma noise. OTOH, humans see less detail in color than brightness. That's why all good de-noisers offer to adjust noise reduction separately and we normally leave some noise in luma, but kill most of it in chroma. Boosting shadow colors would be just the opposite.

Re: low light performance compared

PostPosted: Thu Dec 28, 2017 4:22 pm
by John Brawley
Uli Plank wrote:It's logical that colors disappear close to the lower and upper end of the contrast range.
After all, black is zero color, just like white.


I disagree. Until black clipping, colour saturation INCREASES for the same tone as you pull it down.

Also, Shijan above, with the much smaller and harder to work with Micro does a +6 exposure bracket.

I have shot many times in very low light with multiple copies of the Ursa 4.6 and UMP and it's only in certain circumstances I can see FPN. I've never had to "deal" with it other than usual noise reduction for low light work which usually doesn't work on FPN anyway.

What I do see often is people looking at underexposed LOG footage on a computer screen and crying FPN. But once you apply a grade and look on a TV screen it disappears.

This gets talked about regularly around here.
viewtopic.php?f=2&t=65496
viewtopic.php?f=2&t=61221

I saw the oscar whispered "Disaster Artist" a week ago in a cinema and that thing is swimming in FPN on the screen.

It's a great film to watch about film making and I can't honestly say seeing FPN on the screen took away from the work. No one I went with noticed it.

JB

Re: low light performance compared

PostPosted: Fri Dec 29, 2017 12:14 am
by Jamie LeJeune
John Brawley wrote:I saw the oscar whispered "Disaster Artist" a week ago in a cinema and that thing is swimming in FPN on the screen.

According to Studio Daily it was shot on Epic Dragon with the ISO set to 3200:
http://www.studiodaily.com/2017/12/cinematographer-brandon-trost-creating-disaster-artist/

That explains the FPN. Considering many DP's typically rate RED's sensors below 800 ISO and light accordingly, he was pushing between 2 - 3 stops relative to how it's most often seen on screen.

My personal concern with the two Ursa Mini Pro bodies that I've had so far is that I can't even push them 1 stop before FPN becomes quite visible. In a few cases, it has even appeared as lines in anything blue on properly exposed images. And, in apples to apples comparisons (same light, same settings, same lens) with my original 4.6K, FPN is more apparent when pushing exposure. BMD support reported that these two UMP were both within spec, so apparently my cleaner original 4.6K is a rare "gem" of a sensor that exceeded the spec and I simply got lucky with it.

In an ideal world, I'd simply never push the exposure, but that's not always an option shooting documentary. Becasue the UMP is designed for better ergonomics on doc style work, I expected it to at least have similar tolerance to underexposure as my original 4.6K. Sadly, it just hasn't been the case.

Re: low light performance compared

PostPosted: Fri Dec 29, 2017 8:15 am
by Uli Plank
I'd say they didn't calibrate their Dragon then. I've never seen FPN on a properly calibrated Red Dragon, but lots of dynamic noise at 3200 for sure. But I've not yet seen a camera to deliver clean pictures at that level without noise reduction, be it internal or in post. Maybe Sony rated the new Venice with it's huge sensor so low, just to stay clean when pushing it?

Regarding color in shadows, I admit I simplified. But after all, no electronic camera can deliver pure black. In post I'd normally reduce color in the blacks, if not for stylized looks.

Re: low light performance compared

PostPosted: Fri Dec 29, 2017 12:55 pm
by Benjamin de Menil
My issue with color in blacks isn't with what remains blacks in the graded image. If my original image is very underexposed, and i pull what started as a crush of blacks into the midrange, the lack of color detail is more problematic to my eye than the graininess. De saturating the blacks isn't an option because this is the mid range I'm talking about.

Besides the color vs chroma quality issue, ive also noticed that the dynamic range of the captured image seems to decrease when less light hits the sensor. For instance if i'm looking at a signal that's well spread across the histogram, and i begin to stop down, the graph doesn't just move linearly to the left. It bunches up. I'm curious about the reason for this.

Re: low light performance compared

PostPosted: Fri Dec 29, 2017 2:34 pm
by Uli Plank
It's because such cameras don't see 'black', there's always some signal.

Re: low light performance compared

PostPosted: Fri Dec 29, 2017 4:27 pm
by Benjamin de Menil
Uli Plank wrote:It's because such cameras don't see 'black', there's always some signal.


I don't think that's it. The whole signal compresses - not just the very blackest. Unless maybe it's that the histogram isn't a linear representation.

Re: low light performance compared

PostPosted: Fri Dec 29, 2017 8:45 pm
by Uli Plank
Underexposure will always reduce the dynamic range – just like overexposure.
There is only one level of exposure that gives you the full range.

Re: low light performance compared

PostPosted: Fri Dec 29, 2017 11:07 pm
by Dmitry Shijan
Uli Plank wrote:Underexposure will always reduce the dynamic range – just like overexposure.
There is only one level of exposure that gives you the full range.


If dynamic range of the scene fits to the normal dynamic range of the camera there is no need to underexposure. In most cases bright lights and reflections do not fits to normal dynamic range. So you can clip brightest parts if you prefer digital looking but noise free look. Or you can keep highlights, underexpose and recover some additional grainy f-stops from shadows in post if you prefer more filmic and natural look.
In digital sensors hard clipping can not be recovered. Lot of additional information can be recovered in post only from shadows. But this cause noise and fpn. Filters, fpn masks and downscaling from larger pixel amount to 1080p will remove noise and fpn.
This works well with RAW or Log footage with 10 bit 422 or 444 ProRes captured at native ISO. For those who lazy there are LUTs https://www.thebrim.pictures/wdr.html that designed in similar way and slightly expands the dynamic range.

Do not confuse this method with simple boosting or lowering in-camera ISO and bake it to ProRes footage. Boosting in-camera ISO will NOT expand dynamic range, it will only produce less usable source for post production.

Re: low light performance compared

PostPosted: Fri Dec 29, 2017 11:59 pm
by Dmitry Shijan
Actually you can do the same trick with any crappy camera that shoot jpegs.
1. Underexpose to protect highlights.
2. Adjust Gamma slider to recover shadow information.
3. Add small amount of contrast.
3. Resize to reduce noise and compression artifacts.
4. Desaturate to reduce color artifacts

As example this photo was made with oldschool iPhone 3GS camera:

Image
Image

Re: low light performance compared

PostPosted: Sat Dec 30, 2017 4:12 am
by Benjamin de Menil
Dmitry, very insightful comments. The grade on that iphone image is mind blowing.

From what I've read soft clipping the highlights can also help to make them more filmic. I guess you lose some highlight detail in the process.

This article on the Digital Bolex is interesting.
https://blog.sharegrid.com/blog/inside- ... ital-bolex

Seems like one of its features was a CCD sensor that was tuned to clip highlights more organically. I've shot with it, and to my admittedly untrained eye, it does seem to have more attractive highlight blowout than your average digital camera.