FS700 Noise Test Across Cinegammas

A

AndrewSP

Member
I've been noticing lately that my footage has just a ton of noise, but I have been using Cinegamma 2 at default settings for mostly everything. As nothing I've been shooting with the camera has been very high profile, I haven't messed around with the camera settings beyond their standard values.

Now that I have some time to fuss around with it, I'm a bit concerned about the noise levels across all the Cinegamma (at default settings) as well as the fact that the noise is persistent even at the lowest ISO.

Here are my tests, if anybody can tell me why I'm seeing so much noise... I'd be eternally grateful. Did I end up with a bum sensor maybe, or am I being way too critical?

https://vimeo.com/54684066

Granted, this is all with the AVC internal compression, and I've been considering investing in an external deck the only thing stopping me being the advent of 4K pretty soon. I was thinking a Hyperdeck Shuttle 2 maybe as a stop gap.

I also realize the lightning is sort of low, and I wanted it to be light that to maximize what was recorded in the shadows. I only used an old 4 Bank kino and a 2 bank, both 2 feet.
 
AndrewSP said:
Here are my tests, if anybody can tell me why I'm seeing so much noise... I'd be eternally grateful. Did I end up with a bum sensor maybe, or am I being way too critical?

https://vimeo.com/54684066
Click to expand...

It's very underexposed. This is something a lot of people dont completely understand, but noise is present in shadows of every digital camera ever at every gain setting. If it's not noisy deep in the shadows, the camera isnt designed right and is wasting a lot of dynamic range in the lower end since noise exists to varying extents in the lower end of the signal but there is still a lot of detail there as well. Maybe one day we will have sensors that have 20stops of DR and they will be able to just throw away the shadow stops with the noise in them and give us perfectly clean images with no noise, but until then, you will almost always get the best performance and the least noise by "exposing to the right", and that means possibly using the highest hardware gain (or ISO) setting you can without clipping highlights, putting as much light into your scene as possible, or opening your aperture as much as possible without clipping highlights.

A lot of people make the mistake of thinking low ISO means low noise. Low ISO means the noise starts showing up further into the shadows, but if your image is darker because you are using a too low an ISO, it will still have more noise. In most cases with cameras with hardware gain like the FS700, for example if you use 1stop higher ISO, your whole signal will be boosted by one stop and your noise might start showing up half a stop higher in the shadows, but that means you may have brought 1/2 a stop of your shadows out of the noise. This is a bit of a simplistic description, but you get the idea. The actual tradeoff will depend on the design of the sensor, but for hardware gain (up to a certain point, and some other cameras use software gain past that point) increasing ISO will usually give you less noise for the same scene, right up to the point when its too high an exposure and you have to stop increasing ISO to save your highlights.

Really high ISOs are noisier because they are used to shoot scenes with very low lighting. Noise is worst when you have too little light or when you underexpose your images, but those are not the same thing, since you can have too little light, but you will only make the situation worse by shooting at a very low exposure by setting too low of an ISO value. This is why we have ISO/gain...
 
what Noah said.
neat/MB de noiser2 are your friends. Even the Alexa has quite a bit of noise in the lower end. My C300 was also noisy, and at first gave me the chills, but its totally normal for high DR cams.
 
Noah,

I'm assuming you mean shooting a scene as 'wide open' as you can get away with before you clip your hightlights, then bringing everything down in the grade to get the correct exposure. E.g. If you're shooting s dark scene with no highlights, over-expose the skin tones then bring them back in he grade to where you want them.


I see this makes sense in terms of opening up the iris, but I don't think increasing the ISO will help with this as it just changes the SNR, basically just pulling the whole curve up, bringing the noise with it, so that when you bring it back down in the grade, the noise just ends up right where it started.
 
My FS700 is that noisy, too.
As for Neat Video, it denoises ok with the internal codec but when using my Samurai in Prores HQ I can completely clean up ISO 1600 stuff without any visible loss of resolution (with the temporal denoiser, only, no intra frame denoing). So I can absoluetly recommend an external recording deck !
 
Thanks Noah. I had the same concern as the OP and your post answered all my questions.

I was doing a test comparing the noise levels of the FS700 against my old HPX170 with DVCPROHD codec at around the same exposure level. And yes the HPX170 produces a lot more noise, despite its robust codec (granted they are 2 different cams)

I guess that's why fast lenses are so desirable.
 
Colinelves said:
Noah,

I'm assuming you mean shooting a scene as 'wide open' as you can get away with before you clip your hightlights, then bringing everything down in the grade to get the correct exposure. E.g. If you're shooting s dark scene with no highlights, over-expose the skin tones then bring them back in he grade to where you want them.
Click to expand...

Yes, though in practice, you do have to be careful particular picture profiles settings since you may not simply be able to recover skin tones falling in the knee of the curve, so best not to overexpose skintones too much. The main point here is don'd UNDERexpose if you can avoid it since that is the main source of noise.

Colinelves said:
I see this makes sense in terms of opening up the iris, but I don't think increasing the ISO will help with this as it just changes the SNR, basically just pulling the whole curve up, bringing the noise with it, so that when you bring it back down in the grade, the noise just ends up right where it started.
Click to expand...

That's exactly what I just said does not happen in most cases when it comes to hardware gain. What you are describing is what happens with software gain, which applies a curve after all the noise characteristics of the image have already been baked in. The reality is that in many common modern camera designs (especially those by big manufacturers like Canon, Panasonic, Sony, etc.), the ADC has less DR (more noise) than the sensor itself, so that becomes the limiting factor. That means that applying one stop of gain to the ADC actually may clean up noise in your shadows by nearly a stop (in my first post I said approximately half a stop to illustrate the point, but it depends on the design) while clipping one stop of highlights, meaning SNR is only very slightly changed and your whole exposure range has changed by a stop.

This is well illustrated if you go to dxomark.com and look at nearly any popular digital still camera and look at the DR measurement. As you increase ISO from the lowest level, the DR (ie. noise level) changes very little until it hits a the hardware gain setting that provides the lowest read noise level for the ADC, and after that point, the curve is flat and one stop of DR is lost for every additional stop of ISO/gain applied. I have not tested this in the FS700, but from general use it seems clear that there is a lot to be gained by using gain :)

I'd be interested to see which ISO setting for the FS700 has the lowest read noise. Maybe I'll test it when I have time. I'd guess that even up to ISO5000 the read noise may still continue to decrease.

As I understand it, this essentially the reason cameras like Alexa and BMCC use dual-gain readouts since read noise in the ADC is a common limiting factor in DR. They are able to use an ADC or ADCs that read twice so that their DR is not a limiting factor in collecting the entire signal from the sensor. Presumably RED also operates in this way, as they lock the ADC to the one hardware gain setting that minimizes read noise and they hopefully have an ADC that does not limit DR of the system significantly. None of these digital cinema cameras give the user control over hardware gain because they are designed primarily to deliver the greatest image quality. That means they all may have some part of the sensor's range that remains unused if it is not able to be captured by the ADC. Whereas our FS700s and most other cameras including DSLRs etc are able to use the full range of the sensor by adjusting the gain applied in the ADC so it can capture different portions of the sensor's range, just not all at once, you have to pick and bake in what part of the sensor's DR you want to use. It's a bit of a trade off.
 
chromeboy007 said:
Thanks Noah. I had the same concern as the OP and your post answered all my questions.

I was doing a test comparing the noise levels of the FS700 against my old HPX170 with DVCPROHD codec at around the same exposure level. And yes the HPX170 produces a lot more noise, despite its robust codec (granted they are 2 different cams)

I guess that's why fast lenses are so desirable.
Click to expand...

Actually the irony is that cameras with "more robust" codecs, especially i-frame only codecs, will usually appear noisier. See the test I did recently comparing AVCHD and ProresHQ recorded from my FS700. The prores appears to have more noise in the shadows but also more real detail in those shadows, because AVCHD naturally tends to denoise, since it cannot accurately compress noise or any other such detail that changes randomly from pixel to pixel and frame to frame. The shadow areas that get denoised by AVCHD also tend to lose real detail unfortunately. This is yet another reason to expose to the right and keep your detail out of the shadows/noise if you are shooting AVCHD.

That said, I have no doubt the FS700 is naturally much less noisy at all ISOs than the HPX170 which is older and more importantly gathering about 18x less light at the same f-stop due to its 18x smaller sensor...
 
Last edited:
nyvz said:
Actually the irony is that cameras with "more robust" codecs, especially i-frame only codecs, will usually appear noisier. See the test I did recently comparing AVCHD and ProresHQ recorded from my FS700. The prores appears to have more noise in the shadows but also more real detail in those shadows, because AVCHD naturally tends to denoise, since it cannot accurately compress noise or any other such detail that changes randomly from pixel to pixel and frame to frame. The shadow areas that get denoised by AVCHD also tend to lose real detail unfortunately. This is yet another reason to expose to the right and keep your detail out of the shadows/noise if you are shooting AVCHD.

That said, I have no doubt the FS700 is naturally much less noisy at all ISOs than the HPX170 which is older and more importantly gathering about 18x less light at the same f-stop due to its 18x smaller sensor...
Click to expand...

noah-

thanks for adding to the conversation.

would you please offer a link to the testing you referred to?

thanks in advance.

be well.

rob
smalltalk productions
nyc
 
I do believe that there is something a little odd at times with the shadows on the 700, there seems to be a range of exposure that is noisier than it should be, lower exposure is less noisy as is higher - there's band of noise which is more or less obvious depending on profile. I really don't have the time yet to quantify it but based on observation of hours of footage i seem some situations where i feel there is more noise then there should be (especially compared to the 100 in a similar scene). It might be that the picture profile is pushing a band of exposure in a way that's making it more obvious.

I'm looking at neat video and also the denoise stuff in nuke as well. i haven't decided which way i want to go yet but neat video seems to get peoples vote, although if i'm right (and i've only played with a demo so far) then you have to process the video to see temporal noise reduction? the preview in the app is purely spatial? Any recommended settings for 700 footage through neat?

cheers
Paul
 
robfilms said:
would you please offer a link to the testing you referred to?
Click to expand...

http://www.dvxuser.com/V6/showthread.php?297722

It was a quick test I did out my window in a couple minutes, mostly because I thought a counterpoint was needed to another test I saw floating around that basically shot completely static high key daylight scenes and concluded there is no difference between the quality of avchd and prores... There clearly is a difference which is very content dependent and its importance is subjective depending on the user as with many aspects of image quality, but the difference is there.

paulcurtis said:
I do believe that there is something a little odd at times with the shadows on the 700, there seems to be a range of exposure that is noisier than it should be, lower exposure is less noisy as is higher - there's band of noise which is more or less obvious depending on profile. I really don't have the time yet to quantify it but based on observation of hours of footage i seem some situations where i feel there is more noise then there should be (especially compared to the 100 in a similar scene). It might be that the picture profile is pushing a band of exposure in a way that's making it more obvious.
Click to expand...

I've noticed this to some degree as well. Noise is often there but as with any detail it can be made stronger or made nearly invisible through localized adjustments to contrast. If you bring up your black level and down your black gamma you can make all the noise disappear eventually, but that doesn't mean it's gone, you'll just have milky shadows with such a low slope in your shadow gamma that small details are not discernible from one another. Once you correct your image back to normal, it will all come back, but then again if you want a final milky look then you can get away with a lot more in your shadows. What I'm saying is that black gamma has a big affect on the way noise looks, and because the shadow gamma adjustments on the FS700 are so coarse, I have found that no setting is perfect for maintaining consistent contrast throughout the curve in the shadows. There are certain areas where the noise that happens to have a particular luma value becomes a bit more pronounced, and you can change your black gamma but it usually just shifts the luma value a little so its a different area that looks likely unnatural. That said, it's a pretty slight issue and usually looks worse on the built-in LCD than in final images because the build in LCD tends to clip blacks a bit, or so I find.

I've literally sat with my camera hooked up to an external monitor pointed at a luma ramp just testing lots of black gamma settings and none of them are perfect, but that's just a fact of the camera. Personally I find it's something to be aware of but I rarely have issues with it in my videos. There may be a magical combination, but it is likely a combination of black level and black gamma and it may not be a setting you actually want to use since you'd probably have to correct back it to normal in post. The picture profile options in the FS700 are better than the FS100, but still have a lot of limitations, many of which I think are a much greater issue than the coarseness of the black gamma curve... Most of them relate to color depth, primary colors, fringing, and clipped highlights or even more fundamental IQ issues. Despite these, the FS700 is still ahead of most video cameras as they all have issues like these. It's not so easy to create a gamma curve or a gamma system everyone loves or that works in all situations. Sony has never been great in this regard. Maybe it's because they are still to stuck in the broadcast or consumer worlds.
 
paulcurtis said:
I do believe that there is something a little odd at times with the shadows on the 700, there seems to be a range of exposure that is noisier than it should be, lower exposure is less noisy as is higher - there's band of noise which is more or less obvious depending on profile. I really don't have the time yet to quantify it but based on observation of hours of footage i seem some situations where i feel there is more noise then there should be (especially compared to the 100 in a similar scene). It might be that the picture profile is pushing a band of exposure in a way that's making it more obvious.

I'm looking at neat video and also the denoise stuff in nuke as well. i haven't decided which way i want to go yet but neat video seems to get peoples vote, although if i'm right (and i've only played with a demo so far) then you have to process the video to see temporal noise reduction? the preview in the app is purely spatial? Any recommended settings for 700 footage through neat?

cheers
Paul
Click to expand...

I did this test with my tungsten kit for flickering at 240fps. The results are nothing out of the ordinary, but what I'm most concerned about is the leaf right at 12 o'clock that's mostly in the shadow. Like you are saying it is noisier than I think it should be. I'm new with the FS700 and too bad I didn't test the same scene with another cam but this amount of noise doesn't seem right. What do you guys think?

https://vimeo.com/54338762
 
240fps in low light is always noisy no matter what, much more than other frame rates. In daylight not too bad. Also not sure what you would achieve by comparing the same scene with another camera since the FS700 is the only camera that can do 240fps in its price range
 
We literally got the camera a few days before shooting so there was no way to spend a decent amount of prep with it (and thinking that it's pretty much the same as the 100 with better highlights wasn't ideal)

I think it's very profile dependant and as Noah says, it's all to do with the black gamma. I feel that cinegamma 2 is one of the culprits. Cine2 seems great for low light on the LCD. I think it has a baseline where the gamma is pushed anyway, so if you up the black gamma again then you really are seeing into shadowy bits you shouldn't be. But even so there's a band in there that seems noisier and if that falls on something important it can be a bit annoying.

Noah, pointing it at a luma chart would have been ideal if we'd had time - did you experiment with cine 2? In the end i find cine 4 a little better

cheers
paul
 
Yeah I have noticed that although CINE4 places midtones higher than CINE2, CINE2 has fairly high shadow gamma levels compared to CINE4. I thought I'd be using CINE4 most of the time and CINE2 in low lighting, but I wasn't finding CINE2 to be that much less noisy so I just always use CINE4 (mostly with my nLog profile settings as a base) and tweak from there. I prefer the extra highlight range of the CINE4 gamma even if it pushes the camera a bit, you can always lower noise a bit by using lower black gamma settings. The extra highlight range is critical for me because it helps avoid clipping and a lot of the more serious IQ issues I run into with the FS700 are related to clipped colors or clipping at high contrast edge detail so they can be better avoided by avoiding clipping.
 
Framerates above 60fps are all noisy (and have less sharpness, more moire) because they use line-skipping so they are probably collecting about 2stops less light in a sense. Beyond that, higher framerates necessitate higher shutter speeds so you'll need even more light to compensate.
 
I did some pixel peeping and I noticed that 120fps is decent in low light, worse than 60fps but still OK. However, when you jump from 120 to 240 in low light, the noise increases dramatically and disproportionately when compared to other frame rates. But in daylight 240 looks pretty decent.
 
nyvz said:
It's very underexposed. This is something a lot of people dont completely understand, but noise is present in shadows of every digital camera ever at every gain setting. If it's not noisy deep in the shadows, the camera isnt designed right and is wasting a lot of dynamic range in the lower end since noise exists to varying extents in the lower end of the signal but there is still a lot of detail there as well. Maybe one day we will have sensors that have 20stops of DR and they will be able to just throw away the shadow stops with the noise in them and give us perfectly clean images with no noise, but until then, you will almost always get the best performance and the least noise by "exposing to the right", and that means possibly using the highest hardware gain (or ISO) setting you can without clipping highlights, putting as much light into your scene as possible, or opening your aperture as much as possible without clipping highlights.

A lot of people make the mistake of thinking low ISO means low noise. Low ISO means the noise starts showing up further into the shadows, but if your image is darker because you are using a too low an ISO, it will still have more noise. In most cases with cameras with hardware gain like the FS700, for example if you use 1stop higher ISO, your whole signal will be boosted by one stop and your noise might start showing up half a stop higher in the shadows, but that means you may have brought 1/2 a stop of your shadows out of the noise. This is a bit of a simplistic description, but you get the idea. The actual tradeoff will depend on the design of the sensor, but for hardware gain (up to a certain point, and some other cameras use software gain past that point) increasing ISO will usually give you less noise for the same scene, right up to the point when its too high an exposure and you have to stop increasing ISO to save your highlights.

Really high ISOs are noisier because they are used to shoot scenes with very low lighting. Noise is worst when you have too little light or when you underexpose your images, but those are not the same thing, since you can have too little light, but you will only make the situation worse by shooting at a very low exposure by setting too low of an ISO value. This is why we have ISO/gain...
Click to expand...

That makes a lot of sense actually. I was going about it all wrong. The FS700 monitor seems to be too dark in the lows and too bright in the highlights so I always feel like I am over exposing even by going with the histogram. I guess I should probably take a much needed personal shooting session with the camera to see what I can get out of it.

I'm curious about your nLog vs. the Abelcine set of profiles, especially the "log" version they posted. Where exactly would your two differ in terms of approach and what you're trying to achieve out of the camera versus what they are going with?
 
nyvz said:
you will almost always get the best performance and the least noise by "exposing to the right", and that means possibly using the highest hardware gain (or ISO) setting you can without clipping highlights, putting as much light into your scene as possible, or opening your aperture as much as possible without clipping highlights.
Click to expand...

"The shadow areas that get denoised by AVCHD also tend to lose real detail unfortunately. This is yet another reason to expose to the right and keep your detail out of the shadows/noise if you are shooting AVCHD."

Hi, Noah--

Forgive my lack of understanding, but I'm trying to reconcile the concepts behind these two statements.

In the first, you are recommending exposing as much as possible without clipping highlights (whether via ISO, lighting as much as possible, or opening the aperture). This point I understand...in fact, I often actually overexpose "point" highlights in most of my shots (as long as they are not present in the skin tones).

It is the second point that I'm having trouble understanding...how to keep "detail out of the shadows" if one is also exposing as much as possible (whether via ISO, lighting as much as possible, or opening the aperture). Increasing exposure will bring out detail in the shadows, yes?

What am I missing here? Thanks!
 
You must log in or register to reply here.
Back
Top