MX 2K more noisy than 4K?

[Jon B.]

Has anyone had this experience with the MX sensor?

I was shooting last night in an incredibly dark stage performance and wanted to test slow motion. I set the configuration to 2K 16:9 iso 1600 and immediately saw a noisy blue image. I opted out of the slow effect because it was a live performance (although in hindsight I should have taken a quick grab).

The image was incredibly noisy in the blue channel. It was considerably more noise than with the settings 4K 16:9 iso 1600.

Anyone else with this?

[Gunleik Groven]

There is a simple explanation to this.
When you shoot 4k and downsample to 1080, you also "downsample" the noise (decrease the areal of the noisy pixels) to 1/4th.

When shooting 2k you see the noisy pixels bloom in an almost 1/1 ratio @ 1080.

The downsampling is hiding a lot of the noise.

SO
The cam has the same noise carracteristics, the downsampled image has not.

If you in addition to this shot @ Tungsten, the blue channel will be gained. Try to develop @ 5000 and grade that (it will look yellow). That will probably be less noisy.

Cheers

Gunleik

[Jon B.]

That seems a bit right, although I did some frame grabs today in a quick and dirty test and am having trouble uploading the two images here.

The tifs I grabbed, both 2K and 4K, have a different look in the shadow/blacks. The 2K just slightly more blue noise. I will be shooting again tonight and will grab a couple comparison shots.

[Gunleik Groven]

What would be nice was an r3d grab (from RC-X)

You most host it somewhere else, though. Reduser doesn't allow for files of that size.

(a 4k r3d would be about 1.6 mb)

[Andrew Wilding]

If you punched in on a piece of film, the grain would be enlarged along with the image. The same thing is napping here, but it is a sensor, not a strip of film, and it is noise, not grain. 2k is just punching in and throw g away the excess.

[Jon B.]

I did snapshots in rcx exporting full res tifs. I zoom into the 4K and create the same frame as the 2K. When I view, essentially at the same framed image as the 2K, I see more blue noise. Both r3ds were taken seconds from each other, the difference being 2K and 4K. The colors in 2K are a little more blue.

Too bad I can't post it now, but later if I still haven't figured out the reason, I'll find a way to get these images up there.

I noticed this first under extreme low-light situation so it was more noticeable.

[Gunleik Groven]

Well, I think Andrew and I pretty much have descridbed the reason, and it is in sync with what you see, and actually should expect.

Cheers!

G

[Jon B.]

Shouldn't a 2K inner frame of the 4K grab resemble precisely another 2K grab of the same shot? I didn't realize when shooting 2K that blue noise was boosted?.. this is what I am seeing.

There must be more going on then a cropped sensor.

[Gunleik Groven]

The cropping = 4x the size of the noisy areas @ 2k compared to downsampling 4k to 2k.

Thus, shot @ 2k viewed @ 2k will give more noise.

If you took the 2k image and scaled it down to 1k (1024 x 512 or 576 depending on ratio) with full debayer, the image noise should be comparable to the scaling from 4k to 2k - given that you have a good scaling algorithm.

If you in addition was shooting highspeed, that would lead to a worse compression scenario, probably not decreasing noise...

[Dan Kanes]

Are you running firmware 30.6?

I think it helps 2k ;)

[Gunleik Groven]

On MX, too?

Just curious...

[Jon B.]

Thanks for the replies,

I am running 30.6 and will be looking to get some test shots to post tonight. Granted this is my first shooting with the MX

[Mike Prevette]

Shooting in 2k is roughly similar to S16mm film and 4k is more similar to 35mm. You can't defeat physics. When your shooting 2k you are magnifying a smaller portion of the chip and therefor magnifying the grain more. I would never recommend 2k in a low light situation.

[Jon B.]

I was intending to shoot slow motion and in fact need to use 2K.

Anyway, using the exact same chip to capture for the 4K and the 2K. What I am saying here is that the portion taken out of the 4K image to match the exact frame from the 2K image, the 2K is blue noisy. Yes, the 'grain' is magnifying but the overall colors in the shadows is more blue.

Also, zooming into a 2K frame on the 4K image is also magnifying the pixels, therefore matching the 2K shot. The 2K has more blue in the exact same shot set-up. The only difference should be that the sensor is cropped, not that the blue channel is more noisy.

I will post pics later but I wanted to check if anyone else had this issue in the meantime.

[Michael Olsen]

Increase in perceived noise would be expected (as described above).

The blue shift is curious, though. White balance?

[Jon B.]

Increase in perceived noise would be expected

Right,

just like perceiving more noise in the 4K image when zoomed in. It should match the 2K image when doing this but it doesn't. I zoom into the 4K image to match the 2K image's frame and the 4K is a cleaner less blue shadow; same frame of view.

[Gunleik Groven]

Are you shooting high-speed @ 2k?

[Jon B.]

I had not engaged the varispeed yet and the shutterspeed on both 2k and 4k were the same.

I took some test shots today in daylight conditions and the difference was not as noticeable as last night under tungsten but it was still there.

[Gunleik Groven]

Interesting. I haven't done exactly what you describe.

A fun test would be to shoot 4K-16:9 @ RC 28 & 36 vs 2k @ RC42 under lowlit tungsten...

[David Battistella]

I'm not sure anyone else mentioned this, but in 2K you are not downsampling anything. You are only using the center portion of the 4K imager to capture the images.

This means that 2K is not "scaled from 4K it is only "cut out" of the 4K. So you are using only 1/4 the image area to capture to r3d.

Words that are wrongly used here are "magnification" and "downsmapling" as niether of these things occur when in 2K mode.

By making the image area smaller it allows for higher frame rates and the ability to "overclock" the sensor from what it would be able to do in 4K.

Why does it look like crap in low light. Easy. You are using 1/4 of the sesonr capacity to capture your image, the center area is all that is used. Also you are overclocking this area to get the speed you want to achieve and chances are you are wide open, indoors.

The rule for speed is one stop of extra light needed @ 48fps, 64fps, 72fps, 96fps. That is four stops at 96fps. If you had to "crank" the ISO out to 1600 then you were probably capturing in very low light.

You might probably have been in tungsten lighting as well, whcih is just another barrier to a decent image on CMOS sensors.

The reason daylight looks better is because you are dealing with so much more light.

David

[Jon B.]

Actually the only setting I changed was the 2k from 4k, both at 16:9. Same iso, same shutter. You described it correctly by saying 'cut out' of the sensor. At least that's what I expected. It seems to be processing differently for lack of better terminology on my part.

This is precisely what I did: exported the 4k and 2k tif of the same shot and cut out by zooming into 4k to match the 2k.

Getting some more examples under tungsten now. Will post compressed tifs later to get them on the site.

[David Battistella]

Jon,

A 2K cutout of material captured at 4K may ot macth well because you are using so much more resolution and the full resolving power of the sensor in 4K.

In 2K you are always compromised. I think Mike described it will in the 35mm and 16mm comparision. 16mm just will never be as sharp and has a different grain and noise characteristic.

My general rule is to shoot 3K at 48fps for slomo as going to 2K past about 72 can be very noisy unless you are outside or in some daylight condition.

I also highly reccomend "neatvideo" to clean up the shots. You will be amazed and what it can accomplish on grain.

David

[Uli Plank]

If you shoot 4K and scale down to 2K with any good scaling algorithm, there will be interpolation of neighbouring pixels. That is reducing noise as a side effect.

[Jon B.]

Here is the example,

I recorded a 4K shot, stopped recording, reconfigured to 2K and recorded again.

Afterward, brought the footage into rcx and exported both snapshots to tifs full res. Opened in preview and cropped the 4K to the 2048x1152 2K frame. I then exported both the original 2K tif and 4K cropped (2048x1152) tif to compressed jpegs to attach here.

This is the difference I was talking about between the cropped 4K and 2K... pretty significant in my opinion.

Now, this is my first shoot with my MX sensor outside of testing at home. My question is not what the best way to shoot 2K etc.. I wanted to know if anyone else has seen this on their MX sensors. I feel this is something wrong...

[Jon B.]

If you shoot 4K and scale down to 2K with any good scaling algorithm, there will be interpolation of neighbouring pixels. That is reducing noise as a side effect.

the r3d's look this way.

[Panos Bournias]

[QUOTE=Jon B.;663704]Here is the example,



This is the difference I was talking about between the cropped 4K and 2K... pretty significant in my opinion.

Yes pretty significant, I will check mine to see if it behaves the same...

[Jon B.]

Take a look at the images in post 24,

has anyone seen this on their MX sensors between 2K and 4K? I've only had the camera back a short while and am wondering if this is just my camera?

Before I get the 'why you shootin 2K?' comments please understand this is not about me shooting 2K all the time. I WANT TO KNOW IF THERE IS SOMETHING WRONG WITH MY CAMERA. Although, shooting in 2K for speed ramping the camera should not be looking this way. I do not remember dealing with this prior to the upgrade.

I want to see if there's some info out there before contacting Red about it.

Thanks!

[Stefan Auer]

have you compared all the metadata of those shots?
its hard to tell exactly with the jpg artifacts, but it doesn´t look like there is more noise but just different colorsettings. with gamma settings of R 0.92 G 0.94 B 0.8 on the 2k version it looks about the same to me.

[Jon B.]

No different settings. This was a test shot. I shot 4k, switched to 2K without changing any settings, and even the monitor lit up blue when I did so. Actually that's how I first noticed it.

Trust me when I say that these jpegs look exactly like the full res tifs exported from redcine X, which look exactly like the r3ds.

[Jordi Figueras]

Just thinking out loud here. Maybe I'm wrong but I have a theory about the noise in the Blue Channel with the M sensor and the build 30.5.

That build was released when the M-X sensor hit the street. I worked with it in a M-X camera and find no problem with the image, anyway, when in my (still) M sensor R1 the noise in the blue channel was huge. Then it hit me that the only people in he forum talking about that issue were people without the M-X sensor, so I think that the M-X sensor is far more better in it's blue channel. Hence the less noisy images you can get up 'till 1600 iso, so the 30.5 build was tried over the M-X sensor, so not having problems with the Blue Channel... but people with the M sensor got weird Blue Channel noise and RED helped us with an enhanced blue channel in build 30.6.

Don't know if this makes any sense to the rest of you, but maybe I'd try to do the test but with build 30.5, and see what happens...

Cheers!

[Jon B.]

Thanks for bringing this thread back up. I experienced this with my new MX sensor and build 30.6 BTW.

Went to shoot varispeed, reconfigured to 2K and the screen goes blueish. I think it's a problem with my camera. Waiting on Red to get back to me. If anyone experiences a shift in blue when changing image size please respond.

The difference is so noticeable in the thumbnails the images don't even need to be opened.

reposting...