PDA

View Full Version : Sensor pattern



Felix
09-03-2012, 11:25 AM
Hi,

just messed around with some footage I downloaded and found a disturbing sensor pattern. Not only apparent when brightening up the picture too much but also visible when setting the darkest point of the image to 0 IRE.

Click (http://www.steinhardtverlag.de/Felix/pattern.jpg)


Left side of the image. horizontal lines and weird colors

Here in a much too bright version to emphazise it
Click (http://www.steinhardtverlag.de/Felix/toomuch.jpg)

Iīm afraid, in motion such a fixed pattern will be very disruptive! It was in Franks low light footage, too, but only visible if you brightened it up far too much.

Kholi
09-03-2012, 11:29 AM
Who's frame is that?

Anyway, yes, count on it. Very Underexposed areas will look like this and it can be difficult to get rid of with noise reduction. All the data that you may see in the shadows may or may not be usable because of these factors.

Thing is, all cameras would exhibit this if you were getting the data raw. The less compressed, the more obvious it becomes. Compression cleans things like this up in cameras like the FS100 so and so forth, but given the right amount of underexposure you can see it in those too.

GH2 Hack is much more terrible than this, or at least I think so...

markmwilliams
09-03-2012, 11:36 AM
Possibly that's why JB's advice has been expose for shadows on the BMCC without clipping the highlights?

Felix
09-03-2012, 11:37 AM
I donīt remember were I downloaded the footage. Somewere here in BMCuser. There was also footage of a dog in it.

Hmmm, but if Iīm shooting a low key scene (I donīt mean low light), even if I expose to the right, there would still be dark parts in the image, like in every dark scene from a movie.
So if that means you can never have underexposed areas in your shot (like a Night Skyline) this cam might be good for evenly lit sitcoms... :(

@markwilliams: Exposig for the shadows without clipping highlights to supress noise is ok, but if you have to overexpose your shadows without clipping highlights to avoid sensor pattern, then itīs not 13 stops of DR.

Kholi
09-03-2012, 11:50 AM
There's a difference between severely or "very" underexposed areas and underexposed areas, though. Basically, if you're in a situation where you wish you had more than say 1600 ISO, the you're in trouble.

Here's a place where I dared to rate RED MX @ 6400 ISO:

http://www.superseedsmovie.com/Stills/int_car_lp.jpg

When I get to the original, you'll see just how bad it can get when you don't have proper amounts of light etc. But, even here you can still see the sensor pattern from everything being crazy underexposed.

I mean, the lens was a T3.1 wide open, it's night, and it's being lit by three litepanels and two tiny kino light bulbs the size of your fingers.

If you're in this situation, and you can't tolerate the noise, you've probably chosen the wrong camera. lol

I'll post a raw still from this.

Kholi
09-03-2012, 11:52 AM
@markwilliams: Exposig for the shadows without clipping highlights to supress noise is ok, but if you have to overexpose your shadows without clipping highlights to avoid sensor pattern, then itīs not 13 stops of DR.

That's not correct. You have to do similar (please note I said similar, not the exact same thing) with Alexa because most of its DR is on the top end, not the bottom. Inverse to Epic where most of its DR is on the bottom, not the top. It's still likely to be 13, just not the kind of thirteen for lowlight shooters.

Felix
09-03-2012, 12:03 PM
That's not correct. You have to do similar (please note I said similar, not the exact same thing) with Alexa because most of its DR is on the top end, not the bottom. Inverse to Epic where most of its DR is on the bottom, not the top. It's still likely to be 13, just not the kind of thirteen for lowlight shooters.

Your shot from the red was ISO 6400. What I posted was ISO 800. I didnīt brighten up anything. I even lowered the shadows so the darkest parts are at 0 IRE.

Look, I imagine a classic indoor low key scene with beautiful practicals all over the place, fill light off course, but dark parts, too. ISO 800, no brightening in post, just grading the picture so the shadows are just about to crush.
If that results in a visible sensor pattern...I donīt know.

If itīs true that the latitude is all above middle grey than itīs ok. But if itīs like a, letīs say FS700, you would end up with burned out highlights all over the place because you lightened up the shadows too much.

I have this camera on preorder so I donīt mean to talk bad about it. Just find out if there are problems with it.

Kholi
09-03-2012, 12:09 PM
Your shot from the red was ISO 6400. What I posted was ISO 800. I didnīt brighten up anything. I even lowered the shadows so the darkest parts are at 0 IRE.

ISO doesn't matter. If it's underexposed in RAW it's underexposed. The number I gave was a rating, so basically I could've easily said 800 ISO and I would be right, the entire argument would be thrown out of the window.



Look, I imagine a classic indoor low key scene with beautiful practicals all over the place, fill light off course, but dark parts, too. ISO 800, no brightening in post, just grading the picture so the shadows are just about to crush.
If that results in a visible sensor pattern...I donīt know.

The ISO doesn't matter, it's just a guideline for rating and lighting ratios. And in this scenario no the sensor pattern probably wont' show up if you've rated to say 320~640 (interior) and lit accordingly. You've got JOhn Brawley's DNGs, I don't recall seeing any sensor pattern noise or hatching.



If itīs true that the latitude is all above middle grey than itīs ok. But if itīs like a, letīs say FS700, you would end up with burned out highlights all over the place because you lightened up the shadows too much.

I think the disconnect is what you're used to working with and what working with RAW really means from acquisition to post. It's a little hard to explain, but basically you just don't want anything to be severely underexposed. Just looking at this image that you've posted, everything looks underexposed. You'd have to know what stop it was shot at etc. to know for sure how far under it is.

DanKanes
09-03-2012, 12:18 PM
I posted that as a fov test of the tokina 11-16mm lens

Shot in a dark prep bay at blacklist.

At t4

Http://dankanesbmcc.blogspot.com

Why are you trying to lift the blacks?

Its dark and should be dark. Should be brought down not up.

Let me post a corrected frame for you.

Kholi
09-03-2012, 12:22 PM
Thanks Dan.

Clearly that frame was intentionally shot like that. I don't imagine seeing any FPN if it's handled as it was meant to be.

Felix
09-03-2012, 12:26 PM
I donīt get it completely.
If you went back to Redcine X and rated that shot with ISO 800 the picture would be very very dark and the pattern would be gone.
The BMCC picture is very very dark (also rated at 800) and shows a pattern. Iīm totally fine with such a sensor behavior if I opened up the shadows or the whole brightness to much (like ISO 1600+) but it should not be there at 800. Shadows can be noisy at 800, Iīm fine with that.

I guess underexposed areas just have to be really crushed in post with the BMCC.

@DanKanes: I didnīt lift the blacks I brought them down to 0 IRE. Iīm using a log curve to start with grading and the shadows are much brighter there. Dark should be dark, thatīs correct. Without horizontal lines going through the picture.
Look at the WFM:
click (http://www.steinhardtverlag.de/Felix/log.JPG)

Maybe Iīm just a noob to make it right. I never graded log curves before and for me it just looks like (in your picture from your blog) that you threw out all shadow detail

Kholi
09-03-2012, 12:32 PM
I donīt get it completely.
If you went back to Redcine X and rated that shot with ISO 800 the picture would be very very dark and the pattern would be gone.

It would be very dark, but the pattern would not be gone. It would not vanish, it would not disappear, you're only making an adjustment to the rating. It would all sill be there. But the image was rated at 6400 and lit to 6400, so 800 would not be clean at all, it would be dark and unusable.



The BMCC picture is very very dark (also rated at 800) and shows a pattern. Iīm totally fine with such a sensor behavior if I opened up the shadows or the whole brightness to much (like ISO 1600+) but it should not be there at 800. Shadows can be noisy at 800, Iīm fine with that.

Still off, though. Your ISO doesn't matter. You don't open it up in post, you open it up before it hits the sensor. Lights.

The noise absolutely should be there at 800 if you do not have enough light before the camera.



I guess underexposed areas just have to be really crushed in post with the BMCC.

Still off... nothing has to be crushed, it just has to be captured according to either taste or maximum range...

DanKanes
09-03-2012, 12:42 PM
My thoughts on this frame here:

http://dankanesbmcc.blogspot.com/2012/09/using-raw.html

DanKanes
09-03-2012, 12:45 PM
One note I forgot to add...

Raw records in extended range - so there is actually Sub Zero IRE information recorded.

I am also of the opinion that raw is NOT recorded in LOG on this camera, it's linear but raw.

Can anyone refute that claim

Felix
09-03-2012, 12:47 PM
Ok maybe I just misinterpreted the footage through the log curve!
I just thought it would have the right shadow brightness but obviously it must be darker.

@Dan: I convert the dngs to cineform RAW, which uses 12 bit log curve, definitely not linear.
The camera could be linear, I donīt know.

dustylense
09-03-2012, 12:51 PM
That's not correct. You have to do similar (please note I said similar, not the exact same thing) with Alexa because most of its DR is on the top end, not the bottom. Inverse to Epic where most of its DR is on the bottom, not the top. It's still likely to be 13, just not the kind of thirteen for lowlight shooters.

While I agree with what Kholi has been saying here through the thread, one has to remember that ALL sensors need light. It's digital. The sensor needs something in EV value to record, or it's just gonna say (yes sensors speak), "I don't see anything, so I'm gonna just place my pattern and noise there Mr. Underexposer". Now some cameras that we are used to use software in camera to make that more pleasant. This camera has none of that.
There have been samples posted where they show me that this is not going to be the best night time shooting camera without a bit work in post. I've seen some sensor banding as well as noise. But so what, after you account that you're getting like $1500 in software, you are paying $1500 for this camera. Can't have it all guys!

dustylense
09-03-2012, 01:08 PM
I posted that as a fov test of the tokina 11-16mm lens

Shot in a dark prep bay at blacklist.

At t4

Http://dankanesbmcc.blogspot.com

Why are you trying to lift the blacks?

Its dark and should be dark. Should be brought down not up.

Let me post a corrected frame for you.

I'm happy to have seen that youtube demonstrating the live grade. I too have the MBPR (LOVE IT) and now feel very happy about client confidence monitoring... Thanks for that upload. Also, I agree with you. Blacks are blacks. Why make them grey?

Kholi
09-03-2012, 02:32 PM
Felix, you can download this: www.campcomet.com/raw/Superseeds_5000ISO.zip

I made a mistake, it was rated at 5000 ISO. This won't be a fair comparison for noise patterns because it isn't the same sensor, but as far as the MX sensor goes it's optimized at 800 ISO. So basically, same as the Magic Cam, to get the best results you light/tune/etc. to 800 ISO.

By that standard, this is incredibly far out of spec, and technically this camera was not the right camera for the job.

The Lens is a Cooke 18-100/T3.1, The light sources are three lite panel minis with various color temps (green and orange) and one mini kino car kit (fingersized bulbs) at two different levels. It's dark outside, there aren't any other sources.

Play with it in RCPX, adjust ISO. This is what happens when you don't have enough light to light RAW. On the flipside, turn up noise reduction in RCPX, export the clip at 1080/2K to ProRes 422 HQ

Let me know how that turns out. =]

Quick references:

800ISO (what would be ideal to rate at):
http://www.campcomet.com/stills/Stills/800iso.png

5000ISO (what it was rated at, based exposure on this):
http://www.campcomet.com/stills/Stills/5000iso.png

PaPa
09-03-2012, 04:01 PM
There's a difference between severely or "very" underexposed areas and underexposed areas, though. Basically, if you're in a situation where you wish you had more than say 1600 ISO, the you're in trouble.

Here's a place where I dared to rate RED MX @ 6400 ISO:

http://www.superseedsmovie.com/Stills/int_car_lp.jpg

When I get to the original, you'll see just how bad it can get when you don't have proper amounts of light etc. But, even here you can still see the sensor pattern from everything being crazy underexposed.

I mean, the lens was a T3.1 wide open, it's night, and it's being lit by three litepanels and two tiny kino light bulbs the size of your fingers.

If you're in this situation, and you can't tolerate the noise, you've probably chosen the wrong camera. lol

I'll post a raw still from this.

This is fine. There are countless films where I have seen some scenes with very grainy footage when inside of a car and it's never bothered me. Hell, add a little more noise to break up the pattern.

Kholi
09-03-2012, 04:17 PM
Here's a moving version, highly compressed from the R3D:


https://vimeo.com/48764760
https://vimeo.com/48764760
pw: ssrcpx

Mp4 is downloadable, it's actually pretty decently compressed for what it is. Nothing done to the footage aside from RCPX denoise.

dustylense
09-03-2012, 05:15 PM
This is fine. There are countless films where I have seen some scenes with very grainy footage when inside of a car and it's never bothered me. Hell, add a little more noise to break up the pattern.

And the cool thing is, you can have this all for the LOW LOW price of $1500. (I say $1500 because of the bundled software). Hey, WHAT DAY EH?

PaPa
09-03-2012, 05:53 PM
Yeah, im normally quick to jump in and make purchases like this when great stuff comes out, but for once, I'm going to wait and see what the verdict is after a few months. Make sure i keep up to date with the boards and what not.

Felix
09-03-2012, 06:07 PM
Thank you, especially Kohli for your input!

I wonder what to do when shooting Prores log. I remember reading that you should underexpose log a little because image compression increases with higher brightness. So there should be no skintones overexposed.
I guess exposing to the right is not the right why with log?

Kholi
09-03-2012, 06:17 PM
Thank you, especially Kohli for your input!

I wonder what to do when shooting Prores log. I remember reading that you should underexpose log a little because image compression increases with higher brightness. So there should be no skintones overexposed.
I guess exposing to the right is not the right why with log?

Man, really, I wish I had a camera. Like, it's been a while since I've felt this irritated about owning gear. I don't think it's been this strong since the Letus Ultimate and then the GH1.

I say this because I want to figure this out as well. Here's what I do know, though... transcoding ProRes 4444 from R3D's in a "LOGish" or REDLogfim is a certified workflow for many productions. They adjust the R3Ds to represent the max DR that the image has stored, then dip to ProRes and never look back. You can still do a TON of work with ProRes4444 and even 422HQ 10bit 220mbit files, if they've been processed to the best of their ability. That includes readjusting your white balance if you've botched it.

It's not necessarily the format, though, that you should be worried about exposing for. Unless you're using the "Video" LUT, which looks ... not as pleasant from what I've seen. You would still expose slightly to the right, but you'd need to know where the real clipping point is, and what just appears to clip in camera.

this is where the RGB readout or similar really comes in handy. But the Zebras should do you well.

What we ALL need to see, as a final measure of ProRes internal recording's this: ProRes Internal Recording and External Recording at the same time to a decently priced external device, ProRes external Recording simultaneously as DNG, then the CinemaDNG to ProRes Conversion through Resolve9. Comparing those in LOG space will tell us nearly everything we need to know about the ProRes application.

There's another thread here where I talked about this, but without a camera I can't quantify it. Anyway, I believe that the ProRes internal LOG may actually give you less of a hassle with FPN/Sensor Pattern Noise etc. because it's more compressed from the sensor than say if you spat out ProRes 422 HQ from the CinemaDNG on a computer.

I'm probably wrong, that's all well and good because either way we come out on top, but that's just my theory after checking out the scarce ProRes clips available for download.

All that said, once you've been able to play with the camera and mess around with light in lowlight situations, take the files through the ringer (denoise, color, etc) I think you'll see that it's a lot easier than you expect, but just different from what you're used to with cameras that do everything before it hits the recording medium.


FFFFFF I NEED A CAMERA.

DanKanes
09-03-2012, 06:22 PM
Thank you, especially Kohli for your input!

I wonder what to do when shooting Prores log. I remember reading that you should underexpose log a little because image compression increases with higher brightness. So there should be no skintones overexposed.
I guess exposing to the right is not the right why with log?

Great question.

I personally set the recording gamut to log (aka film) but the viewing gamut to video.

I take the video highlights to just below 100% and your log recording should be preserved.

Another way is to take the log viewing gamut and run it through a LUT box and then shape that signal to what you find acceptable for a first light. Then your log is preserved and you have a rough idea of the transform you want to use for linearization. (see my Livegrade overview for a super basic intro to this workflow).

Felix
09-03-2012, 06:27 PM
Thanks for that detailed answer. I just played a little with Johnīs Prores files and thought about shooting prores because it grades pretty well. The only problem for me is saturation. Log is soooo desaturated that even setting saturation to max inside colorista is not enough...
And you have to do WB and Tint corrections with the colorcurves instead of just WB and Tint settings.

@Dan:According to Black Magic, the 100% Zebras always show RAW sensor clipping. Thatīs why I opened the thread in the official forum asking for zebras from 30-109

Kholi
09-03-2012, 06:31 PM
Thanks for that detailed answer. I just played a little with Johnīs Prores files and thought about shooting prores because it grades pretty well. The only problem for me is saturation. Log is soooo desaturated that even setting saturation to max inside colorista is not enough...
And you have to do WB and Tint corrections with the colorcurves instead of just WB and Tint settings.

xP Time for some Resolve 9. Good thing that it comes with the camera.

You'll start learning about lift, gamma, gain controls and how to work your primaries to prepare for secondaries. And, how saturation of colors can be changed by adjusting contrast, exposure, etc. Vibrance, all that.

Whole new world.

The good thing is? You can copy and paste a look to all of your footage in resolve, few clicks. So if you've saved a few different looks that you love as a baseline image for LOG files coming out of the camera, then you could potentially never have to worry about getting that baseline again.

it's like a picture profile, but you just do it in post instead.

Kholi
09-03-2012, 06:41 PM
Oh I just thought about something... I'm curious how close the video/rec709 viewing and recording gammut is to Alexa's interpretation. They're similar in the way they handle DR, if it's closer to Alexa's than say RED's then that would be interesting.

Edit here: and has ANYONE done any research on ProRes recording changing ISO's and how that affects noise etc?

Felix
09-03-2012, 07:04 PM
And: Has anyone compared resolution of Prores / RAW?

RyGuy
09-03-2012, 07:54 PM
There's another thread here where I talked about this, but without a camera I can't quantify it. Anyway, I believe that the ProRes internal LOG may actually give you less of a hassle with FPN/Sensor Pattern Noise etc. because it's more compressed from the sensor than say if you spat out ProRes 422 HQ from the CinemaDNG on a computer.

Wouldn't the same thing happen with the final output? Say all post was done in cDNG, then exported to a compressed format as most finished projects are... wouldn't the same compression beautification happen, just at a latter step?

When I exported a logish cDNG from Afterglow from resolve to prores 422hq it gave it a very filmic look to it.

Kholi
09-03-2012, 07:56 PM
Wouldn't the same thing happen with the final output? Say all post was done in cDNG, then exported to a compressed format as most finished projects are... wouldn't the same compression beautification happen, just at a latter step?

When I exported a logish cDNG from Afterglow from resolve to prores 422hq it gave it a very filmic look to it.

There honestly should be a difference between a camera with limited hardware doing the conversion from the sensor/DSP than a computer, with a several times the computing power, doing the conversion.

As weird as this sounds, I kinda think the camera's conversion might incidentally do a better job of reducing noise (through compressing bits and data) and NOT preserving these unwanted details than a really good computer would.

But we can't know if I'm crazy or not until someone tries it.

Besides, this is why RED said they don't want to offer ProRes or any other lesser compression as an option in camera: the computer would do a better job.

RyGuy
09-03-2012, 08:04 PM
But we can't know if I'm crazy or not until someone tries it.

Haha I'm not calling you crazy. I think you're onto something, and really look forward to seeing what comes out of this hypothesis.

Felix
09-03-2012, 08:18 PM
As weird as this sounds, I kinda think the camera's conversion might incidentally do a better job of reducing noise (through compressing bits and data) and NOT preserving these unwanted details than a really good computer would.


I tried out all the prores clips from John tonight. Denoising, grading and indeed, they seem less noisy the the RAW files. I really have to compare it for myself when I have the camera.

Kholi
09-03-2012, 08:27 PM
Yeah by default they definitely would be noticeably less noisy. There's supposed to be noise in raw, that's not unexpected, my main get is will it be a little easier (note, easier, not necessarily technically better) to just shoot ProRes in lowlight situations from the get go.

I think John said something to the effect of "no magic sauce", I guess I'd still like to see the results from someone or at least test it myself.

Curiosity that could be easily sated by having mine grubby hands upon this beast.

PaPa
09-04-2012, 10:32 AM
mmm, magic sauce.

IT worked for the big mac, should work here too.