DRAGON vs EPIC MX Noise - NOT GOOD

[Jeff Kilgroe]

These are showing up as 32kb. I think it's just the .rdc folder, not the .r3d file.

Yeah… no R3D files there.

 

[Brian Merlen]

https://www.dropbox.com/sh/paj84ixbx...0Jk7FTZPABIGVa

Here are some Dragon r3ds that feature some under exposed spots if anyone is interested in checking it out

 

[Daniel Stilling]

If memory serves me well, it was said that Dragon at 2000 ISO should behave noise wise like MX at 800.

Its te disappointing to see this.

 

[Patrick Tresch]

Dragon



MX



I don't see the problem with this footage. Developed DragonCoior4 RedLogFilm.

Pat

 

[Ivan McCullough]

I don't see the problem with this footage. Developed DragonCoior4 RedLogFilm.

Pat

RED do

 

[Isaac Marchionna]

I'm glad RED is tackling this issue and so far this thread has been awesome in working through the issue.

 

[Alan Gordon]

In the kitchen shot one thing that I am seeing is that there is a more regular pattern to the Dragon's noise. Something I see often in 5D and something I remember from early Red One. It's also something that I had attributed to an improper black shading but could be something else.

 

[Eric Haase]

Just want to be clear - my intention in posting this thread is to improve the situation. Within a minute after firing up my dragon my gut (and the lowlight goalpost) told me something was not right. I owned a r1 M,MX, epic and now dragon. I just want the best images possible.

Brent if you need my assistance with anything that could be helpful I am willing to help in any way I can. I hope this can get figured out soon.

 

[Patrick Tresch]

RED do

Well, let me rephrase it... MX has more problem than Dragon. The noise is more muddy yelloish on MX.

Pat

 

[Matt Ryan]

Hi guys,

I did a small test this morning. Fresh Black shading. Nikkor 20mm lens (did not even check focus) just did it quick in my office. I am not seeing that clumpy RED channel noise unless I zoom in to about 150-200 percent.

I don't have an MX to do side by side, but basically if you see that noise goal post then the EPIC is telling you that the noise is going to be there. The thing that blew me away the most was that there is a 1 stop noise "threshold" so to speak. It took only one stop (going from 8.0 to 5.6) to eliminate 80% of the noise. That was a surprise to me as I thought it would be much more gradual.

What it means is (and I am talking about how I have to treat my camera today) that if I see that noise goal post, I'm going to to everything to open up or add more light to get rid of it. Of course we all know that in the real world that ain't always possible. If you wide open and the noise goal post is up, it's telling you that there is some noise in that frame.

Underexposure is going to create noise. It's just that simple. As I said in my early findings. Throw light at these sensors and these problems disappear. The first thing I captured on my Dragon was the exact same scene I captured on my MX, a dimly lit church. The dragon far outperformed the MX in that scenario.


My test is so basic.


1. Expose at F8.0
2. Expose at F5.6
3. Expose at F4.0

At F8 the goal post is fully lit-up on the scene. 1:1 crop on the image.

At 5.6 80% of the goal post goes away and so does the noise.

At 4.0 it's that much cleaner.

Here is the link to the R3D's

https://www.dropbox.com/s/aw9ezx7hk68rwuk/3stops.zip

To me, the shot at f4.0 should have no noise at all, none... Nothing is crushed, the noise bar is empty, yet there is noise... This is not right at all...

 

[Elsie N]

Dragon



MX



I don't see the problem with this footage. Developed DragonCoior4 RedLogFilm.

Pat

WOW! Looks good enough for moving pictures, to me.

 

[Gavin Greenwalt]

My theory after doing the dynamic range testing is that it looks like the RED files are cutting off the blacks. So "0" has been calibrated down to -.001 or so. As a result you're seeing a less desirable noise structure where the bottom of the noise floor is below 0.0.

When I did long frame averaged takes I would expect the noise floor to "stabilize" somewhere around 0.0001 instead it stabilized at 0 in some places. That's highly improbable that every single frame of 100+ frames is exactly at 0 (the true bottom of the noise floor).

My super hypothetical theory is this. Dragon used to be an 800 ISO sensor. They 'mechanically' tuned down the analog gain by 50% to make it a 400 ISO sensor like MX. This revealed some extra headroom. But in the process the ADCs are now clipping in the shadows. The fix would seem to be to add a little bit of lift prior to debayering to ensure that no data is getting clipped. .

 

[Matt Ryan]

My theory after doing the dynamic range testing is that it looks like the RED files are cutting off the blacks. So "0" has been calibrated down to -.001 or so. As a result you're seeing a less desirable noise structure where the bottom of the noise floor is below 0.0.

When I did long frame averaged takes I would expect the noise floor to "stabilize" somewhere around 0.0001 instead it stabilized at 0 in some places. That's highly improbable that every single frame of 100+ frames is exactly at 0 (the true bottom of the noise floor).

My super hypothetical theory is this. Dragon used to be an 800 ISO sensor. They 'mechanically' tuned down the analog gain by 50% to make it a 400 ISO sensor like MX. This revealed some extra headroom. But in the process the ADCs are now clipping in the shadows. The fix would seem to be to add a little bit of lift prior to debayering to ensure that no data is getting clipped. .

In theory that should be easy for RED to tweak right? I mean, each stop down low feels like it has a steep drop off to the next stop instead of a gradation from one to the other. Is that a simple way of stating it?

 

[Patrick Tresch]

WOW! Looks good enough for moving pictures, to me.

Just compairing the noise floor of both cameras not saying it's what we need.



I don't see MX being a better camera even if Dragon could be better as advertised (or what did we expect?)

Pat

 

[Elsie N]

Just compairing the noise floor of both cameras not saying it's what we need.



Pat

Understood, but it still pops.

 

[Nate Clapp]

just to add, getting similar noise here. Recently upgraded to Dragon, did a test of Dragon and an MX epic, I upgraded both to current firmware and blackshaded both. Below is a side by side blow-up of a detail- Dragon shot 6KHD with 50mm RPP and Epic-x shot with 35mm 5KHD and physically closer to match frame sizes. Blow up shows more detail in Dragon, but also more noise.....

 

[Rob Macey]

Well,

I realized I was looking at my footage in 1/8 mode in RCX. At 1/2, the images are loaded with red speckled noise even at all the proper exposure and settings. The noise is consistent across each image. It's just more noticeable in the blacks / shadows.

I've re-opened a ticket with RED about it.

Damn...

 

[Brian Robbins]

Did my own quick test today and found similar results.

 

[Patrick Tresch]

To me, the shot at f4.0 should have no noise at all, none... Nothing is crushed, the noise bar is empty, yet there is noise... This is not right at all...

You don't look at RAW but at some RG4 curve set to 800 iso. If you push your low lights (the lowest is the blue channel far left in that picture) to the mid grey (like exposing it to 2000 iso) it's normal you'll have some noise even if the goal post says (at 800 iso) there is no noise. Same applies with big color temp changes (like 5500k to 2700).

Pat

 

[Matt Ryan]

You don't look at RAW but at some RG4 curve set to 800 iso. If you push your low lights (the lowest is the blue channel far left in that picture) to the mid grey (like exposing it to 2000 iso) it's normal you'll have some noise even if the goal post says (at 800 iso) there is no noise. Same applys with big color temp changes (like 5500k to 2700).

Pat

I know how to process footage lol. Switch to Redgamma 4 and tell me you don't see noise...

Look at the R3D yourself and you will see...

Even dropping ISO to 250 still shows some noise (although much less for obvious reasons..)