Canon Community Canon Community
 


Reply
Respected Contributor
Posts: 1,735
Registered: ‎02-28-2013

Re: 7d Markii Grainy


@RobertTheFat wrote:

The apparent success of the 7D2 makes you wonder what we're going to see in the 5D4, doesn't it? I bought a 5D3 this year to begin my conversion to FF. But I guess I won't buy another one just yet.  Smiley Wink


I think the gap between high ISO performance of the 7d2 and 5d3 is more than some internet reports claim.  I don't think that the sensor or processing technology is radically different.  Full frame will always be able to produce better high ISO images than a crop, the physics will never change.  If there really was a marked increase in high ISO performance of the 7d2 system they could simply implement the same structure in a 5d and it would sell.

 

Personally I'm hoping the 5d4 is a high MP camera that allows pixel binning.  And a smaller crop size like the Nikon D810 offers.  Between binning and the Canon high ISO performance I think it'd be phenominal for landscape, architectural, high end portraiture and product photography.

Respected Contributor
Posts: 1,735
Registered: ‎02-28-2013

Re: 7d Markii Grainy


@TCampbell wrote:

I wouldn't blame in on "Lightroom" per se.

 

Anytime you have low signal (dark pixels) and you attempt to bring up the brightness in those dark areas (by any means) you're asking the computer to "amplify" the signal.  But these tools don't tend to accurately discriminate between "signal" and "noise"... the tools tend to amplify everything.

 

It's not so much the tool chosen... just a generalization that "amplifying" signal also tends to amplify the noise along with it. 

 


Oh, I agree with that, and I suspect that's the issue.  I was referring to the post above implicating Lightroom's substandard performance.  But the conversation didn't seem to focus on the fact the image was severely underexposed and pulled up, but more which program was used to do it.

 

I think most can agree that the LR noise reduction isn't spectacular.  I use LR for my RAW conversion and have a default low level (about 20 I believe) noise reduction applied on import just to smooth out normal noise.  But if I have a photo that really has noise in it I export to Photoshop where I use a combination Imagenomincs Noiseware, frequency selection, and manual touchup. 

Esteemed Contributor
Posts: 3,816
Registered: ‎06-11-2013

Re: 7d Markii Grainy

RAW images don't have any color adjustment in them (in part because they don't contain any color "hues" other than straight RGB data).  For example... the camera has "picture styles" in which you can alter saturation, color substitution... or even select straight black & white.  But if you set a camera to black & white and then shoot RAW, the "RAW" image will be a color image.  The camera records the white balance setting only in the meta-data but doesn't apply it to the image.  (post processing software can then apply that white balance adjustment.)

 

At the base layer of the sensor is an array of photo-sites.  These aren't really "pixels" at this level.  Just light-sensitive points which are generally sensitive across the visual spectrum (and beyond).

 

There's no simple way to alter sensitivity of an individual photosite at the time of capture.  Each photosite has a "well depth".  You can think of this as the number of photons it can capture before it's considered "full" and anything beyond that is simply over-flowing (blown out... we don't know how much brighter it was supposed to be because it overflowed.)  If a sensor has a deeper "well depth" then it implies that it has greater dynamic range (it cord more light value differences between simply "completely black" and "completely white").

 

To create a color image, the camera uses a color-filter-array (CFA) and this array is usually a Bayer mask (and in Canon's case it is a Bayer mask).  The mask has a checkboard of green, blue, and red micro-filters that coat the sensor.  Each photosite is behind one of these color filters.  These are arranaged so that every other photo-site is green (50% of them), 25% are blue, and 25% are red.  

 

e.g.

 

GRGR

BGBG

GRGR

BGBG

 

It uses a "mosaic" of colors, but the specific pattern may vary.

 

If you really enlarge an image, you'll notice you don't see the mosaic of solid red, green, or blue squares... you see color hues (blended color).  But that's not actually what the camera captures.  In-camera JPEG processing will de-mosaic the RAW bayer image and process it into color hues.  If you don't shoot with in-camera JPEG and instead use a computer to process the RAW images (e.g. Canon's Digital Photo Professional... or Adobe's Photoshop or Lightroom RAW processing, etc.) then it's actually the computer that de-mosaics the image... the camera doesn't do anything but supply the RAW data (there are no pixels in a RAW file... unless you count the JPEG preview thumbnail which is frequently present in the RAW meta-data).

 

I'm told there are apparently cameras that do "cook" their RAW data.  Canon doesn't do that.

 

Suppose you want to establish the color of the pixel in row 2, column 2 in my sample "array" above.  In reality that's a "green" photosite.. there is no other color.  To "de-mosaic" this and give it a blended color hue, a developer would likely consider the itensity of signal recorded in the photosites surrounding that specific photosite.  You'd notice the levels of red above & below... and the levels of blue left & right... and you might even account for the intensity of the corners.  How much weight you give to each pixels depends on the algorithm.  But when you're done computing the color hue, it becomes a "pixel" in the de-mosaiced version of the image.  It is at this point that you could ALSO apply white balance... if you're light was too warm, perhaps you'd adjust the red component down and the blue component up.  But you really cannot do that until the iamge is de-mosaiced.

 

This means there's actually no opportunity for the camera to apply white balance... it doesn't even apply color.  It just records the proportionate signal levels detected on each of the photo-sites (which are reporting the intensity of level behind a solid color filter.)

 

It turns out there are lots of demosaicing algorithms... if you take the same RAW image and process it through each DPP, Photoshop (or Lightroom) and maybe something else (CaptureOne, Aperture, etc.) and then very carefully inspect the color pixels that you get, you'll find that they aren't actually identical even though they all came from the same original RAW file.  Some people prefer the look of one RAW processing system to another.

 

Also there's not necessarily a "best" algorithm... algorithms that work great for nature photography might not work as well for deep-space astro-images.  Astro-imagers can get a bit fussy about specifically how their RAW data is de-mosaiced.  When I process images via PixInsight (an astro-imaging processing application), I get to select the specific Bayer pattern arrangement (because different cameras may use a different pattern) and I also get to pick the de-mosaicing algorithm from a list of choices offerred (there are more algoirthms than they offer.)

 

---

 

As for amplifying signal and noise.  There are ways to game the system.  When I talk about amplying signal also amplifies the noise, this assumes I am applying a straight "linear" amplification to all pixels.  When we increase the exposure in post-processing software, we apply a linear increase to the pixels in the image.  That would increase everything.  If we use a "curves", "levels", or "shadows & highlights" tool to increase just the dark areas... then we're applying a non-linear "stretch" where we amplify dark pixels more than we amplify light pixels.   Can a developer try to get clever to notice an "outlier" in the data and diminish how much they amplify that particular pixel?   Sure they can.

 

Hence I'm really applying a generalization and not asserting that every program has to do this.

 

But if the program isn't using some mechanism to discriminate then, yeah... it's amplifying everything including the noise.

 

Tim Campbell
5D III, 5D IV, 60Da
VIP
Posts: 11,491
Registered: ‎12-07-2012

Re: 7d Markii Grainy

[ Edited ]

To answer your question the bottom photos are underexposed and therefore can not be judged accurately.  Plus the white balance is off to me.  I really don't see what you mean in the outdoor picture,

You did not mention what lens you used?  Did you use the kit lens from the T1i?

The 7D Mk II can really shine with the best lenses,  It will be no better than most with marginal lenses.

 

Ditch all the add-ons.  Just use you camra and LR and shoot RAW.  Do some more properly exposed tests and try again.

Report back on your findings.

EOS 1DX and 1D Mk IV and several lenses!
Frequent Contributor
Posts: 42
Registered: ‎06-17-2014

Re: 7d Markii Grainy

[ Edited ]

@Skirball wrote:
I was referring to the post above implicating Lightroom's substandard performance. 

I'm not impying (or implicating) it, I'm stating outright that Lr is behind the game when it comes to the quality of its conversions, in direct comparison to competitors like Photo Ninja.

 

Of course the fact that the images are underexposed is highly significant - that's a given. But my main point stands that we're seeing a lot of "expert opinion" flying around in an information vacuum regarding what the OP has actually done to the files; and that - unquestionably - some converters do a far better job of dealing with underexposed files than does Lr.

 

But the fact still remains that until we know exactly what the OP was up to, we're all speculating.

 

One thing is clear though - the 7D Mk II is not "grainy": I've got files at 6400 ISO which are cleaner than the low ISO files the OP has posted up - but not out of Lr. And low ISO files are butter-smooth.

 

Leaping to Lr's defence is pointless, because it might well be responsible for the "problem" here.

 

And it can indeed "add" noise - the Detail sliders in the NR and Sharpening panels are really pretty good at doing just that.

 

Which is why I reiterate - we need to know how Lr was set up for these images..

 

VIP
Posts: 11,491
Registered: ‎12-07-2012

Re: 7d Markii Grainy

"Leaping to Lr's defence is pointless, ..."

 

Yes it is "pointless" because the overwhelming majority of pros use it.  You may not like it or how it handles CR files certainly your opinion and choice  

EOS 1DX and 1D Mk IV and several lenses!
New Contributor
Posts: 3
Registered: ‎12-07-2014

Re: 7d Markii Grainy

I use LR with a Tamron 28-75mm 2.8.  I always shoot RAW.  I have for 5 years now.

 

I talked to Canon and they asked me to use their program.  All I have is night pictures as it is dark when I go to work and when I come home.  I did some test shots and have not had time to upload them yet.  I hope to tonight or this weekend.

Respected Contributor
Posts: 1,735
Registered: ‎02-28-2013

Re: 7d Markii Grainy

[ Edited ]

@KeithR wrote:

I'm not impying (or implicating) it, I'm stating outright that Lr is behind the game when it comes to the quality of its conversions, in direct comparison to competitors like Photo Ninja.

 


That's fine.  And I'm stating outright that I think this is almost entirely in the realm of pixel peeping.  For real world use I don't think there's a significant difference in RAW conversion - when done properly.  And that's why I don't care to nit pick it -the impact of sloppy conversion far, far outweighs the minute differences in RAW conversion programs.  And it takes away a whole lot of the credibility of those arguing for one or another.

 


@KeithR wrote:
Of course the fact that the images are underexposed is highly significant - that's a given. ...But the fact still remains that until we know exactly what the OP was up to, we're all speculating.

 


We do know what the OP was up to, I can see it in the histogram.  The photo is significantly underexposed, a good 1.5 - 2 stops by the look of it.  As you say, it's highly significant.  I don't believe for a second that the choice of RAW converter would have an impact that is anything but negligible compared to the impact of underexposing 1.5 - 2 stops.

 

 


@KeithR wrote:

Leaping to Lr's defence is pointless, because it might well be responsible for the "problem" here.

 


That makes no sense of course.  You don't defend something because it might be responsible for it?  Isn't that exactly when you defend something? This is going to put a whole lot of defense attorneys out of work.

 

The popularity of Lightroom does add to the defence, and it's why I mention it.  No, being popular doesn't make something better.  But in this case it means that there are a whole lot of people using it.    I hear your point, the 7D2 is new and maybe Adobe screwed up in the RAW converter in order to get it out quick.  But with such a large user base I'm confident that if there was a significant amount of noise introduced with the RAW conversion that it would be a known issue by now.

 

Frequent Contributor
Posts: 42
Registered: ‎06-17-2014

Re: 7d Markii Grainy

[ Edited ]

Of course it makes sense.

 

There's been a knee-jerk rush to rule Lr out of the equation because it's Lr, in a complete information vacuum regarding what the OP actually did with Lr: Lr could very easily be the issue here, and immediately to assume that it simply can't be, is what zealots do.

 

Remember that hsbn said:

 

"Lightroom doesn't add noise into low ISO Image regardless of settings. It just doesn't remove it if you turn the settings off"

 

And that's entirely, fundamentally, demonstrably inaccurate. In direct comparison with other converters (and I've done - and still do - a lot of such like-for-like comparisons), Lr's demosaicing algorithm does create a noisier "starting point", with NR off, than other converters; and with NR on, Lr's NR is less efficient or effective than that in some other converters, too.

 

Now let's dial some "Detail" slider in from the Sharpening panel, and even low ISO Lr conversions can be surprisingly noisy. 

 

TCampbell then said:

 

"I wouldn't blame [it] on "Lightroom" per se" 

 

And - again - it was impossible to rule out Lr at that stage of the discussion (and still is), if only because of the demosacing point above.

 

Both of these comments were made in the absence of any pertinent facts about Lr's part in the conversion process, on the assumption that Lr couldn't be the cause of the noise.

 

I'm saying it could, and that to rush to Lr's automatically defence as if the very notion that Lr might be the culprit here was some sort of logical impossibility, is - at the very least - unhelpful.  

 

Just to underline this converter variation: admittedly, it isn't a low ISO image - it's 4000 (four thousand) ISO out of my 7D Mk II - but have a look at this 100% crop straight out of Photo Ninja (at way less than default NR settings) and with no additional PP noise reduction; then look again at the OP's low ISO examples, and then tell me that Lr didn't potentially have a role to play in his disappointing results...

 

(Here's the same file at the image level, just for the Exif: selective sharpening of the foliage in Photoshop, but again, no additional NR).

 

Lightroom can't do this at 4000 ISO - not even close - and it is equally behind the curve, compared to Photo Ninja, at low ISOs too.

 

Take-away point: the OP's problems might indeed be down to Lr, and instantly to go to a default "can't be..." position is, frankly, this.

Highlighted
VIP
Posts: 11,491
Registered: ‎12-07-2012

Re: 7d Markii Grainy

I hope I am not a zealot but I am failing to see what you are trying to prove out of your examples.

The berries are bleeding they are so over saturated. To the point of being blurry.  I see nothing can be drawn from those samples.

Did you use a 7D Mk II and what lens?

EOS 1DX and 1D Mk IV and several lenses!
powered by Lithium

LIKE US on Facebook FOLLOW US on Twitter WATCH US on YouTube CONNECT WITH US on Linkedin WATCH US on Vimeo FOLLOW US on Instagram SHOP CANON at the Canon Online Store
© Canon U.S.A., Inc.   |    Terms of Use   |    Privacy Statement