Epic Low Light Sensitivity

[David Crans]

just curious if anyone has heard anything about the low light sensitivity of the Epic camera? How does it handle grain in low light situations? Same thing goes for the new scarets. Thanks.

[Nano-25]

very good question
i need the answer too

[Tom Lowe]

Which Epic? Mysterium X or Monstro sensor?

One can speculate that there will be significant jumps in sensitivity first with Mysterium X, and then again with Monstro. But it is very, very doubtful that anyone with real knowledge will throw out any numbers at all on this thread. We'll simply have to wait and see.

[Jannard]

There is a big difference between native sensor sensitivity and the ability to shoot high ISO in low light. Some sensors have huge pixels and very high sensitivity... but not necessarily low noise. Others have less native sensitivity and very low noise, which allows them to be "pushed" into high ISO. The big advantage to Mysterium-X is lower noise, which allows you to shoot higher ISO with the same noise levels as a much lower ISO.

Jim

[Stefan Christou]

Jim since your on this thread can I ask you to please ditch the idea of sticking ASICs in Epics and Scarlets and just ship em with FPGAs and whatever code you've got at the moment. Please. Or at-least, throw us a bone.

[Justin O'Neill]

Jim since your on this thread can I ask you to please ditch the idea of sticking ASICs in Epics and Scarlets and just ship em with FPGAs and whatever code you've got at the moment. Please. Or at-least, throw us a bone.

I would definitely prefer ASICs even if it takes longer. There are way too many benefits of ASICs over FPGAs.

[Stefan Christou]

Well you know, better late than never, but I'm getting inpatient. I just guess I'll have to get over it and keep shooting. After all, it's the subject not the camera.

[Eirik Tyrihjel]

I would definitely prefer ASICs even if it takes longer. There are way too many benefits of ASICs over FPGAs.

I am in deep water here (I am no technical guy - really), but as far as I understand the benefit of ASICs is it´s dedication to it´s task, and therefor their short boot time - but after shipping there is very few options to upgrade (bad), while as with a FPGA you have a long boot time (bad)- but you can continue to develop the thing for as long as it exists... (very good!)

I must say I don´t like the long boot time on my RED ONE, but i LOVE the way RED has kept improving, and keeps improving. (through firmware uogrades)

IMHO The FPGA seems to fit RED´s nature better, I would hate to have an EPIC camera with glitches that couldn´t be fixed, or needed to be shipped back for fixes.

I have learned to live with the long boot times, whenever I am DIT, I use every natural break in the shoot to swap batteries, and usually nobody ever notices, especially now that the RED sends out a monitor signal within a few seconds of booting.

[Tim Whitcomb]

Jim since your on this thread can I ask you to please ditch the idea of sticking ASICs in Epics and Scarlets and just ship em with FPGAs and whatever code you've got at the moment. Please. Or at-least, throw us a bone.

new policy going to prevent a prompt answer here... wouldn't make sense
for them to comment. as much as Im sure they would like to :)

Lots of new astuff right now to get familiar with and master.

all in good time and remember, in this recovering economy
"All Delays are Divine"

[Jannard]

A little bit if info can lead to dangerous assumptions. Of course we can upgrade the firmware on EPIC & Scarlet. Never said we ONLY had an ASIC in there...

Jim

[Blair S. Paulsen]

ASICs should be a big improvement in so many areas: faster boots, higher throughput, a more refined image for monitoring, higher frame rates, etc. While it could make certain types of upgrades more problematic, the upside is huge.

In terms of sensitivity, I would expect the ASICs to help lower the noise floor and in so doing improve usable sensitivity. Time will tell if I am right about that.

Cheers- #19

[Patrick Scheller]

Thanks Jim for those insights. Here is a link to a page which talks about the differences between FPGS vs ASIC nowadays. They made progress too you know?

http://vlsifaq.blogspot.com/2007/11/what-is-difference-between-fpga-and.html

And here some product info on the latest FPGA chips from the market leader which can easily run a Linux!
http://www.xilinx.com/products/virtex6/index.htm

[Eirik Tyrihjel]

A little bit if info can lead to dangerous assumptions. Of course we can upgrade the firmware on EPIC & Scarlet. Never said we ONLY had an ASIC in there...

Jim

Music in my ears!
My RED ONE #476 is the best thing I ever bought - it almost makes more money than I do...
Togheter we´re dynamite!

[Brook Willard]

Any word on gain, gentlemen? I know it's not always ideal, but there are definitely times where adding a few dB would help immensely. I really hope there's a gain circuit in the camera. If I'm an idiot for wanting it, I'm open to hearing why. :) It just feels like a little analog amplification could help in some situations.

[Jannard]

There are many more features and controls in the next gen. cameras. We are shooting 5K footage now in a test bed. The footage is incredibly clean. There is at least 1.5 stops less noise. We have a lot of controls in the R1. We have a lot more in EPIC and Scarlet.

Jim

[Raphael Varandas]

Holly F... Great words Jim...
Waiting to see those little controls

Thanks

[jbeale]

1.5 stops improvement in noise is very impressive! If you have just one stop less noise, that means you are getting the same image performance with half the total light. If you are starting with a camera that is already very good (R1) it's quite an achievement... I'm sure there are many who are eager to put this claim to the test.

[Jared Caldwell]

1.5 stops less noise...music to my ears. :}

[jbeale]

Any word on gain, gentlemen? I know it's not always ideal, but there are definitely times where adding a few dB would help immensely. I really hope there's a gain circuit in the camera. If I'm an idiot for wanting it, I'm open to hearing why. :) It just feels like a little analog amplification could help in some situations.

If the noise level from your sensor + analog stage is already significantly above 1 LSB on the A/D (as it appears to be on R1) then additional analog gain would not improve your image quality. It would only reduce your dynamic range. Increasing gain in post (digital gain) would give you the same end result without the dynamic range tradeoff.

By contrast, if your analog signal is so clean that you can see quantization steps, then more analog gain would give you more real information than could be achieved by digital gain in post. But again, I don't believe that is the case, at least on R1.

[Casey Green]

There are many more features and controls in the next gen. cameras. We are shooting 5K footage now in a test bed. The footage is incredibly clean. There is at least 1.5 stops less noise. We have a lot of controls in the R1. We have a lot more in EPIC and Scarlet.

Jim

It seems to me that every obstacle that comes up along the way, Mr. Jannard is personally seeing to it that they are obliterated... one by one.

Fascinating.

[Roberto Lequeux]

1.5 stops! Thank you so very much for the info!!

So let me see if I can get this straight with my limited technical knowledge:

1.5 stops lower noise would mean that we'd go from ISO 320 to ISO 800 correct? Also it would mean that we could stop down from f/3.3 to f/5.6 and still use the same amount of light that would be necessary for a Red One with equal DoF, right?

I wish I could find the post where David Mullen answered this question, but since I can't, could someone please tell me the approximate stop down necessary to get the same DoF you get with 4k when you want to shoot 5k?

[Andrew Wilding]

Wow. Exciting news. Incredible.

[david farland]

.....could someone please tell me the approximate stop down necessary to get the same DoF you get with 4k when you want to shoot 5k?

I’ve based my examples on CoC of 0.2mm (5 lines pair/mm at 250mm).....'a' limit of normal human perception.
Note it's below the limits of other Coc effects. Also you're mainly interested in sensor size (diagonals) not resolutions as Coc's is a radial effect unless constricted by other properties.
But comparing the R1 sensor horizontal (24.4mm) to the Epic X sensor horizontal (30mm), the CoC's calculations become…...

RedOne
250mm /24.4mm = 10.25
0.2mm /10.25 = .019 (CoC)

Epic X
250mm /30mm = 8.33
0.2mm /8.33 = .024 (CoC)

Epic FF35
250mm /36mm = 6.94
0.2mm /6.94 = .029 (CoC)

So maintaining the same DOF/FOVs, the equations (don't forget lens crop factor 0.8) are……

R1/50mm/f5.6 / 2.0 metres = .343m (depth of field)
EpicX/62mm/f6.8 / 2.0 metres = .343m (depth of field)
Epic FF35/74mm/f8 / 2.0 metres = .341m (depth of field)


f5.6 to f6.8 gives a 1/2 stop reduction (from RedOne) due to Epic X sensor size increase.

I'm not sure if 1.5 stops of noise equates to an increase of 1.5 'usable' stops of dynamic range in the Epic X, but compared to the R1 you may have another stop which is kinda what Red announced last year they would achieve with the Epic s35, although they probably meant 1 absolute stop not considering maintenance of DOF. The biggie is will they achieve another 2 stops with the FF35!

Dave

[Roberto Lequeux]

Dave, thanks for running the numbers. I really appreciate it.

So (you might laugh) what was the answer? :) Was it 1/2 a stop?

[david farland]

lol...... yes I am.
The answer after all that junk is you now have another 1 stop over the R1 (for the same DOF experience!)
Dave,

[Roberto Lequeux]

The answer after all that junk is you now have another 1 stop over the R1 (for the same DOF experience!)
Dave,

:arf: Are you fucking shitting me?! So pulling focus can be the same as on a Red One, only with higher resolution, wider filed of view at equal focal lengths, and you only need half the light?!?!

Am I understating this correctly?? Cause if that is in fact the case, boy o boy... wow...


lol...... yes I am.

BTW, I am glad to bring laughter to ya. :thumbsup:

[david farland]

Most of what you said is correct, except for FOV......it stays the same if you choose a slightly longer (x1.25) lens. And yes, on a larger sensor with same size pixels, you will get higher resolution. In this scenario, if the Epic X sensor is 1.5 stops faster than the RedOne's sensor and you lose 1/2 stop due to maintaining DOF on a larger sensor, then the result is you have 1 extra stop to play with.
D

[Roberto Lequeux]

Well, yes, if you chose a slightly longer lens you will get the same, but if you put an 18mm on a Red One and then you put it on an Epic-X shooting 5k you get a wider FOV than on the R1.

[Patrick Scheller]

@Jim: If I understand this right, then with 1,5 more sensitivity on the sensor it could also mean 1,5 more on the dynamic range. Can you confirm or disproof that Jim?

[Andrew Gentle]

By contrast, if your analog signal is so clean that you can see quantization steps, then more analog gain would give you more real information than could be achieved by digital gain in post.

If that were the case, it would be time to change the bit depth on the ADC!

[Jannard]

@Jim: If I understand this right, then with 1,5 more sensitivity on the sensor it could also mean 1,5 more on the dynamic range. Can you confirm or disproof that Jim?

They are related, but not all the same.

Engineering DR is defined as signal/noise. So technically DR has increased. DR is also defined as the difference between, or ratio, of the lightest and darkest elements on a printed or displayed image. Close, but not exactly the same.

Sensitivity can be discussed as native sensitivity or useable sensitivity. If you are completely happy with the noise level at ISO 800 of a sensor with a native sensitivity (optimum) of ISO 200, compared to a sensor with native sensitivity of ISO 200 but only yields acceptable noise results to ISO 400, is the 1st one more sensitive? Technically no. Practically yes.

The common ingredient in the discussion is noise. Lower the inherent noise level of a sensor (as we have done with Mysterium-X) and you can push it further with better results.

Additionally, some companies have gotten pretty good at NR inside the DSP. That is not the same as inherently less noise from the sensor. You pay a price for noise reduction after the sensor.

Jim

[Andrew Wilding]

You pay a price for noise reduction after the sensor.

Jim

I wonder if thats part of the reason I find 5dmk ii footage so "plasticy." It all looks like its been run through some rather strong noise reduction software. I guess it could be just the compression on the image though, I don't know.

Im beyond excited for these new Red sensors. The improvements that came with build 20 have already scrambled my eggs, i you will. I cant even imagine what the future holds.

[Jacek Zakowicz]

You pay a price for noise reduction after the sensor.

Jim

To simplify things noise reduction is just "chopping off" certain frequencies(in the upper range) where the noise usually "resides"
Ufortunately these are usually the same or (or close to) frequencies that contain the finer detail of the image (resolution)
So, generally speaking, the price for noise reduction is resolution and the skill is to find the proper balance when the image is relatively noise free and resolution loss minimal. This is where the individual taste interferes, unfortunately....

[Lauri Kettunen]

You pay a price for noise reduction after the sensor.

... I quite can't understand why some manufactures have noise reduction within the camera instead of leaving it into post. Of course, for immediate viewing purposes NR is ok as long as it is not hardwired to the data file.

It's bit surprising that camera manufactures still employ ISO settings in digital cameras. For, that's all history now. Consequently, for me it would make more sense to have just raw metering showing when one hits the right end and then for viewing one could adjust stops, such as -1 1/3, +2 2/3 etc. telling the difference between sensor exposure and viewing. In low light conditions this gave immediately an idea how many stops (or bits) from the maximum dynamics (bit depth, resp.) are lost. The point is, the camera should encourage the user for the best practice --shoot to the right-- resulting in the maximum image quality.