horrible noise on URSA mini 4.6k (?)

I had a URSA mini 4.6k on rent last week, the overall quality looks quite good at 60fps, but when setting it to 100fps (or 120fps) there's a horrible (fixed pattern?) noise that makes the footage completely unusable.
Is this something known? or is it a problem on this certain camera?

The higher frame rate is reducing exposure, which is likely putting your image in the low-light zone where noise becomes more apparent.

The higher frame rate is reducing exposure, which is likely putting your image in the low-light zone where noise becomes more apparent.

nope, definitely not... it was outdoor, bright sunshine and the lenses used were Zeiss and Sigma f/1.4
so absolutely no danger for "low light"...
And the image doesn't look like noise is "more apparent" it looks like noise was the main content on the image

maybe I can get some original footage footage tomorrow so I can post it

Yes, when you increase the frame rate Alex, you still need to open the Iris by 1-stop from 60fps to 120 foster to compensate the exposure back to what is was at 60fps. Even when outdoors, changing the frame rate to a faster one is like stopping the lens down. You can underexpose ca camera in any lighting situation.
You can also increase ISO, but that adds noise.
DS

Yes, when you increase the frame rate Alex, you still need to open the Iris by 1-stop from 60fps to 120 foster to compensate the exposure back to what is was at 60fps. Even when outdoors, changing the frame rate to a faster one is like stopping the lens down. You can underexpose ca camera in any lighting situation.
You can also increase ISO, but that adds noise.
DS

yes, I've been working for 11 Years as a professional photographer now, I know I have to open the iris... and I also know that the image shouldn't look like recorded with ISO 6400 when changing the framerate from 60 to 100...

Sorry Alex, I was just trying to make sure you realized changing frame rates changes the exposure.

There are a lot of new users on this forum that do not understand this. It is hard to tell them from more experienced users like you by looking at the question. So I have to make some basic assumptions, based on how the question is asked.

Personally, I have more than 50 years of experience as a photographer, shootting cameras from a 8x10 Deardorff to a 16mm news film camera, including Bolex S16, and made a living as ENG TV cameraman wi both film and video (ENG type) cameras. I have also been involved in several Cinematography productions, off and on, over the years. Saying you are a "professional" only means you are making money doing it, nothing to do with your knowledge or skill level. That that's sorted, I am only here to try and give back some of the tips/tricks I have learned over the years, and try to help less experienced or upcoming photographers, as I can. I did not mean to insult you Alex.

The Urs Mini 4.6 is a new camera, so we are all learning g about it together. So now back to your question/problem.

The fixed pattern issue on BM cameras usually is the result of underexposure. Hence the wrong assumption about exposure. But, it was a starting point. A little more information would be helpful in sorting the issue. What codec were you recording, ProRes, Film Log, Raw or? Are you shooting full frame of the sensor 4.6K or a windowed resolution at 60fps before switching to 120fps?

I remember reading about issues with shooting at 120fps, which forces the camera in a widow mode to get any frame rate greater than 60fps, which also might be contributing to the issue. I know other users have successfully shot the Ursa Mini at higher frame rates, and this was one of its selling points. Perhaps a UM 4.6 user, like John Brawley or Captain Hook, who have far more experience with UM 4.6 than anyone else here, experience in Shootimg the UM 4.6 can step in to give you more help with this. The additional info will help them to answer your question, hopefully.

Shooting over rank at higher frame rates is not always as strait forward as it should be, as you probably already know. I am done here, good luck.
Cheers

Alex, you're not crazy. Denny, you probably don't know what he's talking about.

Alex i experienced the same thing with my ursa mini 4.6k. I have proof.

I just finished a video where I did two shots at 120fps. Denny, i properly exposed it as I know slow motion requires more light.

at 0:27 seconds in my video

you can see horrible crosshatch and noise, even though i tried noise reduction. Seems it's just part of the image. I liken it to when viewing an image in photoshop and you zoom in all the way to the max you can see the lines between the each sensor.... node.. or whatever its called.

This does seem like a huge problem, my camera is pretty noisy in general. It's nothing like the raw footage they released when they released the camera.

Andrew, we are discussing two different issues here. Your footage sample is not the same as the Alex's question. Looks like your sample footage was shot at ISO 1600, with a 360-degree shutter, given the bleeding around the car lights. In this situation, you are going to have more video noise, it's unavoidable.
Also,shooting at 100/120 fps the camera goes to window mode at 2K or HD, which is going to,result in a smaller sensor sampl of the image, which adds noise, especially in low light like this.
Given your shooting situation, your video came out fairly well, most people viewing it on the web will not be aware of the issues you see by pixel peeping. The bleeding of the tail lights and the off color around the headlights looks worse than the video noise. And, I do know from whence I speak (most of the time), my friend.
Cheers

I'm out and about so can't really do any true pixel peeping on the shots here

But

We do know that 120fps is a compromise right ? I think of the Ursa Mini as a 60fps camera that can do 120 in a pinch...if you really need it, you have a compromised 120fps.

It has to be compromised because...you're using about 25% of the sensor compared to 60fps.

Anyone who's shot Red will have experienced the same thing. Once you crop that much it starts becoming more and more of a compromise.

I'm shooting a series now and I was pleasantly surprised at how well the 120fps tests looked. Sure they were kind of "gritty" but the post house actually said they looked way better than any RED cropped windowed footage they'd seen.

I would still reluctantly shoot 120fps only when I really need it and when I don't have much choice.

If you're thinking 120fps isn't going to be something of a compromise compared to 60fps full sensor then you should adjust your expectations.

Of course if there's something that is more technically wrong that I'm not seeing them forgive my post.

JB.

I can confirm what JB says about shooting slo-mo on a Red.
Once you window down to 2K (300 fps on Epic) it becomes blurry and noisier. The UM46 looks better in perceived resolution, since Red has an OLPF. But noise becomes more apparent when you can't supersample any more (and a Red can have quite some noise too…).

if you look at Andrew's footage here at full resolution you can see really bad patterned distortions in the crowd of people wearing dark clothes. It's not blurriness or aliasing.

Not surprised Kyle, given the frame rate, shutter speed and low light shooting situation. Have you done a frame to frame peek at high-speed film shots? Lots of artificsts their also. With ENG CCD cameras from 20 years ago, the car lights would be a long smearing trail. Is the BM 4.6k camera perfect? No, but it does give more options in shooting than most. High speed overcranking is going to be a compromise on any camera not specifically designed for high speed shooting, like the Varicam. Can you get useable results? Yes, but depending on how you are going to use them in post production.
Cheers

Alex and Andrew: I suggest taking a look at the samples posted in the forum thread on crosshatching to see if it matches the problems you're seeing at 120fps. The crosshatching shows up most often when the timeline resolution matches the sensor size. I noticed it on my camera the first time I shot at 120fps because it was the first time I had set my Ursa Mini to record a 1080 window and put it on a 1080 timeline. If you post a raw frame from an affected shot on the crosshatch thread, others with the issue can download it and perhaps see if it matches the same issue we've been seeing. If you do have crosshatching, BMD has been issuing RMAs to fix the problem.

Not sure if this would help, but I'm shooting 4.6K or 2K windowed and non-windowed on a 2K or HD timeline with HD deliverables and rarely have a problem on my URSA Mini 4.6K PL. I try to use the best quality parameters along the way.


Sent from my iPad using Tapatalk

thx Jamie,
yes this crosshatching issue seems pretty much to be what I got...
I haven't been to the studio today, so I couldn't get any sample, I hope somebody will be there at the weekend, maybe they can send me something.

Just as it was asked: I recorded ProRes422 HQ - Film. It was absolutely bad in the original footage and it got horrible after putting in some contrast in the grading...

Have a look at this thread and see if it looks like what you are seeing.

viewtopic.php?f=2&t=51936

at 0:27 seconds in my video you can see horrible crosshatch and noise, even though i tried noise reduction.

Hi Andrew,

I tried to find what you're describing at 27 seconds, but I honestly can't see what you're describing.

I do see a shot of a car heading away to camera left of the camera. The focus is set to the people standing beyond where the car is, probably unintentionally. I also see evidence of a grade process that has caused the specular highlights and high chroma highlights to clip in a very ugly and not usual way. This looks similar to a now infamous problem that was "leaked" from early UM4.6K footage of planes at an airport at night that had been mis-processed in post with the older BM colour applied in Resolve.

Aside from being( probably?) being mis-focussed at the long end of a zoom with a bit of chromatic aberration on highlights and blooming on the road markings (again signs of a poorly performing lens) and the usual grittiness of a windowed sensor that's only 25% of the size of the full sensor shooting at night under artificial lights....I guess I'm not seeing it....

Its hard to pick noise on youtube under these conditions, but it doesn't look aberrant for this camera to me. I don't see any "cross hatching" either but again, maybe it's not showing up for me in YouTube.

As a general observation, your grade is also very "lifted". There's nothing that's hitting true black or even near true black in your footage. While it's fine to make that choice for a look, it does mean noise will be revealed very readily. A quick measure for me is to look at the black of the image on the surround (same in a cinema). The grade looks very "loggy", but with a lot of pushed chroma in the top end causing those hard super saturated colours like the tail lights to seem very off.

I do see what I'd call aliasing on some highlights as they move across the field, something I would expect on a windowed sensor like this.

I've never really relied on the windowed sensor of a RED nor the Ursa Mini 4.6K, though at least because it doesn't have an OLPF, the windowed crops tend to hold up better on the UM4.6K.


jb

Screen_Shot_2016-10-15_at_1_21_31_pm.jpg (266.72 KiB) Viewed 10676 times

Crop_Screen_Shot_2016-10-15_at_1_21_31_pm copy.jpg (176.79 KiB) Viewed 10675 times

Dr. John, is it time for you to host a weekday noon hour talk show for DPs diagnosing all kinds of ailments and prescribing the medicine needed to get a film back on its sprockets and running again? Maybe something to do when you retire!


Sent from my iPad using Tapatalk

Hey John (and everyone else too),

I'm not meaning to take a side as to whether what we are seeing is reasonable to see, whether for technical reasons like underexposure, or simply for fair arguments of price.

Just to help the discussion, the weird patterned stuff is visible in that shot on the crowd where they are wearing black. It's also only visible on YouTube when played back on the highest quality setting, 2160p. The pattern and resulting interference patterns are invisible on YouTube at smaller resolutions, because the underlying pattern in the image is lost in YouTube's transcoding for resizing.

Also, how easy it is to see may depend what size the window you view it on, since the patterning/aliasing effect is evident at different sizes when viewed with different scaling percentages because the resulting alias pattern is the remainder of the division between the clip size, and the scaled size.

In this image it looks to me like the "crosshatching" issue from that other thread, in which it seems that offsets in brightness levels between the odd numbered pixel rows and the even numbered ones cause small maze patterns that then in turn cause artifacts in scaling on some playback platforms. People in other threads have identified various ways to fix this in the camera, or in debayering too.

My BMCC has a slight difference between the top of the frame and the bottom of the frame, they all potentially do. It has to do with how the sensor is designed and read. You only see the difference on dark material. Some have it more than others, and I suppose if there was a budget for it, there could be some kind of custom calibration to equalize them. I wonder if the 4.6k sensor isn't read odd lines vs even lines, and if that has something to do with what we're seeing here.

At least one person got their RMA'd camera back and the problem was gone after custom factory adjustment. Short of custom balancing each camera, there is also a way to balance green pixels in debayering, but right now only some RAW image software debayering algorithms include it. The fact that they were able to remove the problem in a generic way that isn't custom for each camera gives hope that a firmware fix for in camera debayering to ProRes and an update for Resolve to amend that debayering method are forthcoming.

At that point, this may become a problem with the camera that only affects software debayering RAW images in applications where the debayer software doesn't include a green pixel balancing algorithm, which right now is most of them?

Hey John (and everyone else too),

I'm not meaning to take a side as to whether what we are seeing is reasonable to see, whether for technical reasons like underexposure, or simply for fair arguments of price.

Just to help the discussion, the weird patterned stuff is visible in that shot on the crowd where they are wearing black. It's also only visible on YouTube when played back on the highest quality setting, 2160p. The pattern and resulting interference patterns are invisible on YouTube at smaller resolutions, because the underlying pattern in the image is lost in YouTube's transcoding for resizing.

Hello Kyle.

So here's what I can reproduce after your suggestions.

I can see a grid in the shadows when looking at 2160 in a smaller window. But I can only see it at 2160 in a non 1920 window.

This is NOT FPN.

2160_reduced.jpg (209.31 KiB) Viewed 10536 times

This is the same frame displayed at 1920 full screen within YouTube's player. The grid doesn't show.

2160@1920.jpg (350.33 KiB) Viewed 10536 times

And for variation, here's the 720 version, as it appears in my browser (no grid for me)

720_reduced.jpg (174.19 KiB) Viewed 10536 times

That to me says it's not about Youtube's compression hiding it, it's more to do with the scaling you choose.

This seems very similar to the pocket issue, as already discussed. I personally don't see this happening much because most of my own personal work is displayed at 1920 (Television). And I have seen this issue on the Sony F55.

For reference here is the extremely long thread on Sony's forums. It looks like the exact same issue. When I cam across it in my use of the F55, the same ideas applied. It shows up in preview monitors that aren't scaled to your viewing size, but on final render to the standard resolutions (like 1920 displayed as 1920) it disappears. http://community.sony.com/t5/F5-F55/F55 ... d-p/201051

From memory too with the pocket, BM did release a firmware update that improved the "grid" issue in one of their earlier versions of firmware. I've never had the problem there either, but once that firmware update went out, it seemed to stop coming up as an issue as much. I have also seen the grid happen on the display of the BMCC2.5K in certain flare situations, but it never makes it into my finished work, hence the constant discussion about scaling.

To me, and I don't have a lot of personal experience with this issue, it looks like a similar problem that pocket users have been dealing with (and Sony!). There's this thread (I just did a search on pocket grid) viewtopic.php?f=2&t=12961&start=50#p155873 and in it they eventually talk about V1.8 offering an improvement to the grid like artefact. The other workaround also involved doing a 1 pixel offset in Resolve. Has anyone tried that ?

The only difference here is that the grid in the pocket (and Sony) was introduced most easily on flares (in the blacks) whereas this is happening in non flare situations.

jb

I agree that it isn't FPN, it IS a scaling issue.

But the reason it's a scaling issue is that there is a repeating fault in the source image that becomes aliasing in resizing. I also saw BM took an RMA camera and returned it and then it didnt have the issue, and I also saw it go away in debayering when a setting about the green balancing is adjusted.

So I do expect that a forthcoming firmware might well fix it, by incorporating the same fix into the BM debayering algorithm.

FWIW I do still see it in your 1920 window, but the "grain" is very small.

Thanks for the info. I will take some time and try to digest the Sony link.

Enclosed finally the screenshots from ungraded original footage!

I agree that it isn't FPN, it IS a scaling issue.

But the reason it's a scaling issue is that there is a repeating fault in the source image that becomes aliasing in resizing.

A scaling issue shouldn't be possible, as the higher frame rates only are recorded in a windowed mode, where the pixels should be recorded 1:1

Quite the contrary, the UHD footage, that indeed is resized from 4.6k to UHD, is good quality!

Enclosed finally the screenshots from ungraded original footage!

That looks way to over-sharpend to me.

Enclosed finally the screenshots from ungraded original footage!

That looks way to over-sharpend to me.

absolutely not sharpened!
original untouched footage right out of the cam!
enclosed, for comparison, footage recorded on the same cam, with 60fps, 1080p full frame mode... (so scaled)

Alex, on the last Kopie.jpg screenshot, I'm not seeing any problem. Is that correct from your perspective?


Sent from my iPad using Tapatalk

Enclosed finally the screenshots from ungraded original footage!

I can see it. Your camera is also suffering from the infamous "crosshatching" issue due to an imbalance in the debayering process. Something is not properly balancing the two green channels of the bayer filter, and therefore a grid gets imposed on the image whereby all the odd rows/columns are a different average brightness than the even rows/columns.

Several people have been researching this including myself. My findings were mentioned here: https://forum.blackmagicdesign.com/viewtopic.php?f=2&t=51936&start=150#p305096

Here's an infographic highlighting the "bayer green split" issue in one of your images (and what could happen if it needed to be resized).

The CFA green imbalance issue is actually fixable at the firmware level before the DNG is written, but it can also be done in post before demosaic is performed. RawTherapee has a feature called Green Equilibration that is essentially a green channel imbalance correction. The simplest way to fix it is just a weighted sum between a center pixel and surrounding green pixels in a small region.

If that even/odd problem is the root of the problem, surely it will be fixed by BMD either at the sensor level or the firmware processing the signals.


Sent from my iPad using Tapatalk

If that even/odd problem is the root of the problem, surely it will be fixed by BMD...

It appears to be easy to fix. Only BMD knows where it fits into the camera pipeline, whether its firmware or calibration. The fact the CornerFix and ExifTool remedies are so simple makes me doubt it's a hardware problem.

The challenge is hoping BMD recognizes that many of these disparate complaints and RMA requests are caused by the same underlying issue. People see the symptoms under different circumstances, so they describe the problem differently to BMD support. Some see the "crosshatching" on their computers, to which BMD responds it's a display scaling issue. Others call it "fixed pattern noise", and BMD says that they don't see it (because it isn't FPN). But ultimately it's the same debayering problem cropping up in several threads as the root cause.

I actually think this issue is more widespread than anyone realizes. One person sent their camera in on an RMA only to get a replacement with the same issue. I'm sure there are many other people with the problem who haven't noticed it yet because they haven't triggered one of the secondary symptoms.

So, I hope Blackmagic is piecing this together. That's part of why I'm lending my voice to the discussion even though I don't even have one of these cameras yet.

Enclosed finally the screenshots from ungraded original footage!

That looks way to over-sharpend to me.

absolutely not sharpened!
original untouched footage right out of the cam!

So you shot in ProRes?
Any chance you had the post sharping (meant for live video feed) in the camera set to ON?

Have a look at a Red Scarlet, it has noise too at 1600 and needs light. And they seem to have users…

I pushed the 5D Mark III iso at 6400 and...no noise!!! You can belive this?

Yeah bigger sensor and a pretty aggressive realtime/onboard denoiser...

What I can't understand though... if you're typical project is something where you need to shoot with ISO 6400 on a 5D Mk3... why would you get an Ursa Mini in the first place...

Yes, why not a Sony A7S?

What I can't understand though... if you're typical project is something where you need to shoot with ISO 6400 on a 5D Mk3... why would you get an Ursa Mini in the first place...

First at all, i was buy ursa mini because i belived in Blackmagic revolution promises..."Cinema for everyone" is a a very enthusiastic slogan and i was fooled. Second, i was see short movie with ursa mini and i like it his cinematic aspect in light condition. Plus that ursa mini can record internal RAW, and this is a great advantage. A7SII is great in low-light but look like a video camera aspect, not cinema, and can't record RAW. 5D mark III have more cinematic look and can record internal RAW with Magic Lantern. I hope i was clear.

Enclosed finally the screenshots from ungraded original footage!

I found that too. Was really worried. Took all of the footage to a very important post house and they got the same look when they imported it into Resolve. Then they exported and everything was fine. I don't know how it got fixed but apparently Resolve knows how to do it.

We also checked how noisier 120fps windowed video was. Turns out, it has exactly the same amount of noise as a full sensor 24fps footage. The thing is, we don't notice it much because its "scaled down" to UHD or HD depending on you monitor/output. They were really surprised, as with RED it is much much worse.

(on a sidenote, we also tested the camera against the Sony FS7, with internal recording and prores external, then the FS5, then Scarlet Dragon. It came out as a clear winner in almost every aspect, with the Dragon sensor being close in some)

I tested the UM46 against the Epic Dragon and it was better in some parameters and worse in others, but damn close. Still, "cinematic" doesn't come out of the box, it needs a lot of experience and work in post too.

So you shot in ProRes?
Any chance you had the post sharping (meant for live video feed) in the camera set to ON?

Yes, ProRes!
I can't check it any more, but I don't think so, as the footage recorded in UHD and 1080p FullFrame doesn't look sharpened

Enclosed finally the screenshots from ungraded original footage!

I found that too. Was really worried. Took all of the footage to a very important post house and they got the same look when they imported it into Resolve. Then they exported and everything was fine. I don't know how it got fixed but apparently Resolve knows how to do it.

We also checked how noisier 120fps windowed video was. Turns out, it has exactly the same amount of noise as a full sensor 24fps footage. The thing is, we don't notice it much because its "scaled down" to UHD or HD depending on you monitor/output. They were really surprised, as with RED it is much much worse.

(on a sidenote, we also tested the camera against the Sony FS7, with internal recording and prores external, then the FS5, then Scarlet Dragon. It came out as a clear winner in almost every aspect, with the Dragon sensor being close in some)

I was grading the footage in Resolve, and tried to denies it directly in there... Helped, but what has been still there was some kind of noisy grid structure... (image enclosed) I didn't try to export un-denoised footage to be honest, but I can't imagine Resolve does it on it's own...
Did you shoot RAW or ProRes? If RAW it could be Resolve corrects the problem in the debayering but I shot ProRes, so it's already debayered and fix in the file

Looks good on my computer.

Looks good on my computer.

Do you mean Alex Green's golf image? Here's some detail at 300%:

Bildschirmfoto-2016-10-20-um-14.23.52-Kopie.jpg (19.29 KiB) Viewed 9808 times

Again, this problem is easy to miss when viewing at 1:1, but it rears its ugly head when scaled off-ratio. A problem at the micro level then becomes a problem at the macro level.

Yep, zoomed all the way in and it disappeared.

Yep, zoomed all the way in and it disappeared.

That's a bit hard to believe, unless what you're using is blurring the JPG.

Again, this problem is easy to miss when viewing at 1:1, but it rears its ugly head when scaled off-ratio. A problem at the micro level then becomes a problem at the macro level.

It's also clearly visible at 1:1, especially as the grid stays while the image itself is moving...

My bad, see it now. Time for an RMA.