DRAGON vs EPIC MX Noise - NOT GOOD

[Stacey Spears]

Here is a clip from the Xyla-21 tests that Gavin and I did. I provided ~5 frames of each, so you can run ADD if you like. The left couple of steps on the MX are masked off. I could not find a similar one for Dragon where we did not use frame averaging. To be clear, frame averaging was not used on either of these clips.

1. Noise level on MX is similar for red and green channels and noticeably worse for blue. To be expected given the light source.
2. Noise on dragon is lowest in the green channel. Red noise is higher, close to blue.
3. I set both clips to ISO 2000 and both are on RC3 / RLF to make them as equal as possible. Of course it is metadata, so you can change it.
4. I white balanced on one of the steps. I don't remember which one.
5. I think the noise on Dragon is finer than MX, but you can see it is reddish.
6. Matt, I figure you can post images as you are good at that sort of thing. If you do, you should post two Dragon's, w/ and w/o ADD for completeness.
7. The test clips http://www.spearsandmunsil.com/sspears70/red/06_09_14/epic_noise.zip

 

[Dunkan Wolf]

Tonight I did a side by side with both cameras- Epic MX and Dragon.

1) Both cameras were freshly updated to current release build 5.1.51
2) Both cameras were brought to 65 degrees, set in Adaptive fan with target 65 degrees and blackshaded with port caps on and dark fabric over port cap as well. Blackshade was saved to USER HS and the I made sure to select USER HS and press SET.
3) Cameras both shooting with Leica R 19mm at a 2.8
4) Both cameras at 6:1 compression
5) Dragon at 6KHD, MX at 5KHD
6) ISO set to 800 on both

At the bottom of this post is a link to the R3D files (10 frame trims of each shot) A001 is MX, A006 is DRAGON. I have also included photos of the cameras' touch screens to demonstrate how the goalposts and exposure tool behaves differently on MX and DRAGON, as I documented in a previous thread found here: http://www.reduser.net/forum/showth...POSTS-and-Exposure-Tool&p=1376074#post1376074

You can draw your own conclusions and FWIW here are mine:

1. (My) Dragon is noticeably noisier than MX in the lowlight areas, given same firmware, exposure, ISO setting, temperature and proper blackshade.

2. Dragon noise is different than MX in color and contrast.

3. Goalposts light up full on Dragon and only halfway on MX given the same scene. Exposure tool, however and conversely, shows less purple on Dragon and more purple on MX given same scene.

4. I studied some clips that a friend provided shot on Dragon with the old OLPF- very low light scenes. Those images were MUCH cleaner. Phil Holland's original low light images with old OLPF were MUCH CLEANER.

5. I am very disappointed with the upgrade and I hope this can be resolved very quickly. I tried to test this as thoroughly as possible given I am only interested in comparing the noise in the lowlights and tried to have the fewest amount of variables possible. I hope others can do similar tests. I'd love to see a solution to this. For me, now, it is a real issue. I'll be shooting everything at the lowest possible ISO to fill the sensor with light until that fix is reached.

I am more than happy to work with anyone from RED on this.

***************DOWNLOAD R3D TRIMS and touchscreen photos HERE: https://www.dropbox.com/s/ayb9wvkwvy4ju6y/R3D TRIMS NOISE TEST.zip

The 'OLPF update' was actually a sensor revision

 

[Ivan McCullough]

The 'OLPF update' was actually a sensor revision

If you do not know this to be a fact do not post it.

What is your source for this?

 

[Nate Clapp]

That left image looks SUPER compressed. Like 15:1 vs the Dragon's more lightly compressed image.

Both were 8:1. They are side by side tiff made in red cine. No advanced dragon debayering. Just straight snapshot for both. Then zoomed in quite a bit for detail comparison. I'm away from my computer or I could show full frame.

 

[Matt Ryan]

If you do not know this to be a fact do not post it.

What is your source for this?

I believe he should have substituted sensor revision for sensor re-calibration.

 

[Matt Ryan]

Here is a clip from the Xyla-21 tests that Gavin and I did. I provided ~5 frames of each, so you can run ADD if you like. The left couple of steps on the MX are masked off. I could not find a similar one for Dragon where we did not use frame averaging. To be clear, frame averaging was not used on either of these clips.

1. Noise level on MX is similar for red and green channels and noticeably worse for blue. To be expected given the light source.
2. Noise on dragon is lowest in the green channel. Red noise is higher, close to blue.
3. I set both clips to ISO 2000 and both are on RC3 / RLF to make them as equal as possible. Of course it is metadata, so you can change it.
4. I white balanced on one of the steps. I don't remember which one.
5. I think the noise on Dragon is finer than MX, but you can see it is reddish.
6. Matt, I figure you can post images as you are good at that sort of thing. If you do, you should post two Dragon's, w/ and w/o ADD for completeness.
7. The test clips http://www.spearsandmunsil.com/sspears70/red/06_09_14/epic_noise.zip

 

[Rob Macey]

Shot some int.'s just now at 250 iso in a darker room. Although the noise is less than at 800, it is there. On Epic MX at 250 the black and shadow levels, in dark interiors, absolutely stayed noiseless.

 

[Elsie N]

Hmmmm... this suggests to me that RED could tweak their debayer algorithm and give up a stop or two and get cleaner remaining stops.

 

[Matt Ryan]

Hmmmm... this suggests to me that RED could tweak their debayer algorithm and give up a stop or two and get cleaner remaining stops.

That's basically what the old calibration with old olpf was. Much cleaner low end but a stop less up top. The new olpf/calibration has a stop more up top but at the expense of losing a stop on the low end, which means more noise on the low end.

 

[Gavin Greenwalt]

Both were 8:1. They are side by side tiff made in red cine. No advanced dragon debayering. Just straight snapshot for both. Then zoomed in quite a bit for detail comparison. I'm away from my computer or I could show full frame.

Looks like you found one of those magical times when REDCode bombs hard. Were there trees and grass somewhere in frame?

 

[paul schefz]

That's basically what the old calibration with old olpf was. Much cleaner low end but a stop less up top. The new olpf/calibration has a stop more up top but at the expense of losing a stop on the low end, which means more noise on the low end.

either way.....it is a long way from clean 2000iso...unless you take into consideration the noise in the shadows and expose accordingly, which means you are blowing out highlights...
for the most part i don't mind shooting at 250iso but i was really hoping for at least clean 1600iso or a stop over what i would go to with my epic...iso 800.....which in reality (according to my light meter) is more like 400 anyway....

i don't see how any of this can be changed with firmware?

 

[Brian Merlen]

even if its not firmware and needs another OLPF change its kind of too late for me to not have one...i like red and i don't mind rolling with the punches and am sure it will be worked out whether its firmware or new OLPF or what have you... in the end it will all work out i am sure...

 

[Rob Macey]

I ran the fans in "adaptive" at 65 and "auto" at 45 to see if heat had any effect on the noise. I didn't see any difference.

 

[Thomas Koch]

Can someone test this and report back?

On my Dagon I noticed today that the noise floor (as viewed at iso 12500) goes up pretty dramatically as you decrease the shutter. Ie at 90 or 45 degrees and decreases at shutters more than 180 degrees. I had an idea to do a black shade calibration at 45 degree shutter at 23.97 fps and then see how that compared to the black shade done at 180 degree shutter. In fact the calibration done at 45 degrees shutter is a lot quieter in terms of visible noise (again viewing at the extreme iso of 12500) even at the normal 180 degree shutter. Should this be expected? any downfalls to this?
Thanks
Taylor

 

[Matthew J]

I am glad I saw this because I was getting ready to send mine in for the new OLPF in a week and considering what and how I normally shoot, this would have been a big deal. Here's an example of the old OLPF that's in my Dragon and how it looks totally different in terms of noise in the dark areas: http://www.mrdop.com/dragonoldolpf.zip

Matthew

 

[Matthew J]

On my Dagon I noticed today that the noise floor (as viewed at iso 12500) goes up pretty dramatically as you decrease the shutter. Ie at 90 or 45 degrees and decreases at shutters more than 180 degrees. I had an idea to do a black shade calibration at 45 degree shutter at 23.97 fps and then see how that compared to the black shade done at 180 degree shutter. In fact the calibration done at 45 degrees shutter is a lot quieter in terms of visible noise (again viewing at the extreme iso of 12500) even at the normal 180 degree shutter. Should this be expected? any downfalls to this?
Thanks
Taylor

Are you opening up to compensate for the light loss when you go to a higher shutter speed?

 

[Taylor Morrison]

Matthew,

Observing noise floor with a capped camera body only. No lens attached. I am still finding this trick to work but only with slight improvement. I am fairly certain this is a characteristic of the Dragon sensor as I never observed this with my MX. Please check it yourself as I would like to know if others are seeing this too. First test is to do a normal black shade at 180 degree shutter and 23.98FPS. Make sure its loaded in and then raise ISO to the extreme at 12500 (please note I am in no way expecting to use the camera at this ISO…just a way of observing the noise characteristic of the sensor) As an additional step enable focus peaking for a more dramatic view. Now with the sensor capped simply adjust the shutter speed to 90 degrees or 45 degrees or even less leaving the frame rate alone( I use degrees but I think its 1/48 that equals 180 degrees at 24fps so smaller shutter would be 1/90 or 1/180th…correct me if i am wrong here). I experience more observable noise each step down in shutter speed. Conversely, less noise as I go the opposite direction towards a more open shutter.
My trick is to black shade at 45 degree shutter and shoot at a 180 degrees which does seem to help a bit with noise, but by no means cures all that ails. I know this is not SOP by hey I'll take what I can just now.
Let me know what you find

Cheers

Taylor

 

[Matthew J]

Matthew,

Observing noise floor with a capped camera body only. No lens attached. I am still finding this trick to work but only with slight improvement. I am fairly certain this is a characteristic of the Dragon sensor as I never observed this with my MX. Please check it yourself as I would like to know if others are seeing this too. First test is to do a normal black shade at 180 degree shutter and 23.98FPS. Make sure its loaded in and then raise ISO to the extreme at 12500 (please note I am in no way expecting to use the camera at this ISO…just a way of observing the noise characteristic of the sensor) As an additional step enable focus peaking for a more dramatic view. Now with the sensor capped simply adjust the shutter speed to 90 degrees or 45 degrees or even less leaving the frame rate alone( I use degrees but I think its 1/48 that equals 180 degrees at 24fps so smaller shutter would be 1/90 or 1/180th…correct me if i am wrong here). I experience more observable noise each step down in shutter speed. Conversely, less noise as I go the opposite direction towards a more open shutter.
My trick is to black shade at 45 degree shutter and shoot at a 180 degrees which does seem to help a bit with noise, but by no means cures all that ails. I know this is not SOP by hey I'll take what I can just now.
Let me know what you find

The issue I see here is that each time you increase the shutter speed, you are pushing the sensor further and further. So yeah, you should see more noise if the light amount stays the same (in your case, none) because the sensor is being pushed harder. The difference between ISO and shutter is that the ISO is just pushing the RAW data up, whereas the shutter is actually changing the physical readout speed on the sensor. The noise is going to be that much more noticeable when the photosites have less time to take in light and the amount of light hasn't increased to offset the speed. Does that make sense?

 

[Nate Clapp]

Looks like you found one of those magical times when REDCode bombs hard. Were there trees and grass somewhere in frame?

here is an example of full frame. can't do actual still as that is too big for wordpress site (!) so this is a screen grab. But gives you an idea of how zoomed in i am.

http://nateclapp.com/wp-content/uploads/2014/06/screen-shot-dragon.png

 

[Mark Toia]

i beg toia to post a few r3d's to see if its all that different or if we are just pixel peeping geeks.

I haven't got the new OLPF in David... So my images are fine.

I'm as curious as you are. I'm sitting on the fence until the boys at RED come back with a solution... Which I'm sure they will.