November 29, 2014, 02:49:29 AM

Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - jrista

Pages: [1] 2 3 ... 312
1
http://www.dpreview.com/articles/0892989376/real-world-test-going-pro-with-the-samsung-nx1


Nice! :D I really love this camera! I think it's pretty much settled...good EF adapters or not, I'm getting an NX1.

2
That's my point. Statistically, the NX1 has lower noise, which leads to richer color and contrast (hence the reason the parts of the NX1 image that are supposed to be dark look dark! :P )

It does have very slightly lower noise, but that is not the reason it's a darker image which is what leads to the difference in our perception of color and contrast.

Again I'll note that with color NR the 7D II ends up looking a tiny bit cleaner, i.e. lower luminance noise. But in the end neither requires a different work flow or more work. Neither has "better data."

My perception is a bit different and I tend to agree with  Jrista.
   My comparison of RAW samples (no NR at all - no Chroma no Luminance  NR)  at DP comparison tool   which I posted a bit earlier shows that at  ISO3200 NX1 RAW samples visually  look better/cleaner/crisper than ones from 7Dm2  and as I also noted NX1 noise pattern is visually better and more pleasant than noise pattern of 7Dm2. NX1 is less blotchy , has a bit more higher noise frequencies and a bit more regular  - so should be easier to clean up compared to 7Dm2 . Blotchy noise pattern from many Canon Cameras is something that was always irritating  me as well as low performance at low contrast details in red channel - just smearing them away. 1Dx fortunately is better in this respect though also suffering a bit.   
Also on this samples NX1 is not downsampled/normalized  to 7Dm2 resolution which would also add more difference in favor of NX1.

  I was interested to see if NX1 could compete with A7S - but so far A7S is far ahead of all the competitors.
Just see here  as an example one of  test picture done handheld by A7S in extremely low  light  conditions at ISO20000.
One small screen snapshot and also  full image exported from LR


Thanks for the analysis and image samples. I think you nailed it on the head, and it's the same stuff I've been saying for a while now. Noise characteristic matters. It isn't simply the amount of noise, it's how the noise presents. I am right there with you...I really dislike the blotchy color noise of Canon cameras. Give me clean, random, primarily luminance noise, and even if it is slightly higher, I'll be much, much happier. Gaussian luminance noise distribution is so much easier to clean up, and it required minimal application of NR.


One thing regarding the A7s...it is a 12mp camera. I would be very curious to see how the NX1 performs relative to it when the image sizes are normalized. I don't think the NX1 will be better, it is APS-C afterall, and the total light gathering capacity is still limited, despite the BSI design (which, now that more information is out, may actually NOT be an ISOCELL design...not sure why, guess they figured the pixels were big enough to not suffer enough from color crosstalk to warrant the ISOCELL?) I do think, though, that when normalized down to 12mp, the NX1 will approach the A7s's IQ more closely. 

3
EOS Bodies / Re: Canon 7D Mark II - DXOMark Review
« on: November 28, 2014, 02:51:31 PM »
From Roger, the respected sensor reviewer of Clarkvision:

"A note on DXO numbers. It appears that DXO is not correcting Nikons truncating of the raw data, which artificially improves dynamic range by about a stop. Also Nikon filters the raw data, improving noise and dynamic range further. I believe, based on some experiments, that if the canon data were treated similarly, it would result it numbers at least as good.

Previous to the 7D2 and 6D, pattern noise was a real limitation in Canon cameras (the 1D4 and 1DX are also pretty good, but not a good as the 7D2 and 6D). That pattern noise produced poor shadow areas compared to what could be extracted from Nikon sensors, especially at some ISOs (like 200 and 400 on many canon cameras). But Nikon's raw files look "wormy" in the shadow areas from the in camera filtering of the raw data. The Canon 7D2 raw data looks much more random, as it should be." -
http://www.naturescapes.net/forums/viewtopic.php?f=57&t=249565&start=112

True words. I remember having heard it somewhere that Nikon's RAW isn't actually RAW but somewhat preprocessed. I wonder if their "lead in DR" which is often claimed here, origins from this preprocessing.

And I have another link regarding the 7DII's sensor performance: http://www.clarkvision.com/reviews/evaluation-canon-7dii/index.html


Hate to say it, but this is simply not true. Clipping to the black point in a Canon file would cut out a lot of data, and DR would not meaningfully improve (it might improve numerically, but there are caveats). Clipping marginally improved DR at best. A group of Nikon hackers actually removed the black point clipping in Exmor cameras a while back. In their testing, the difference in terms of noise with the bias offset restored was minimal...at most, ISO 100 RN was 6e-, and in some models lower than that. That is in comparison to the base RN of around 3e- in Exmor-based cameras at ISO 100. That resulted in less than a stop loss in DR overall, sometimes less than half a stop (depending on the camera).


That still means that a bias offset restored Exmor camera still has around 13 stops of DR, to a Canon cameras 11 stops. Operating in exactly the same way...with a bias offset rather than black point clipping. Why so little change? Because Canon cameras as much as 25-40 electrons worth of read noise at ISO 100...that is orders of magnitude greater than what Exmor has. You would have to "clip" a very significant amount of the signal in a Canon file...so while DR may improve, instead of discarding ~3e- worth of potentially usable data, you would be throwing away 22-37e- worth of potentially usable data. That's a LOT of data...that's a lot of potentially recoverable shadow data that is now simply GONE. Now...it's barely there to begin with in a Canon camera, because the read noise is so bad. So, maybe it doesn't matter so much that the information is clipped out.


However, you have to recognize what's actually happening to the signal. Canon's DR numbers may improve...but they are not improving because of a reduction in read noise (which is what happened with Exmor sensors), thereby increasing the amount of recoverable data. They are improving because of a removal of information, thereby DECREASING the amount of recoverable data.

4
Some more information on this:


http://image-sensors-world.blogspot.com/2014/11/active-pixel-color-sampling-story-goes.html


The commentators still seem to think it's a hoax, given the discrepancies and errors in all of the data. Take it for what you will.


I just...thought I'd quote this, maybe make it stand out more. So far, guys with WAAY more knowledge than any of us here regarding sensor technology think the whole thing is a cleverly deployed hoax. It could be real...but, these guys usually don't claim hoax unless there is something really wrong with the information... :P

5
Bummer they are using a 16bit ADC...I guess with S-LOG they are basically shifting the shadow tones up in-camera, so in some ways they could be benefiting from that 21 stops...which is fine for video. For stills, they should really be using a 20-bit ADC. I guess we'll see if they stick this sensor in a stills body or not, but if they do...I certainly hope it would use 20-bit ADC.

Ideally, a 21-bit ADC, to be pedantic.


Indeed. :P

6
According to PixInsight, the noise on the black part of the Fiddlers Elbow bottle, just under the label, has a standard deviation of:


7DII:  0.048 (R), 0.048 (G), 0.051 (B)
NX1: 0.037 (R), 0.037 (G), 0.038 (B)


Statistically, the NX1 DOES have lower noise.

Are we sure that Samsung isn't cooking the RAW images—either with a denoise algorithm or by exposing to the right and then scaling the values?


Well, I know for a fact that the 7D II is cooking the RAW with DIGIC 6. :P I mean, that's what DIGIC 6 does...cook data. :D If it wasn't for that, I highly doubt the 7D II would be able to achieve the IQ levels it does.


I don't know about the NX1...however, given that the current trend IS to cook the RAWs, I have to assume so. Nikon and Sony both cook their RAWs...Nikon with Expeed and Sony with Bionz X. Canon just got into that game. The NX1 sounds like it has the most powerful processor ever to be placed into a camera, and if they weren't utilizing that processing power to reduce noise, I'd be pretty surprised.

7
Quote
To see if my feelings about the images were correct, I saved the cropped areas as 32-bit TIFF and ran both through PixInsight's statistics tool.

You're splitting hairs. No one would ever know the difference in a processed print.


We keep getting to this point, and I keep saying the same thing, and my point always seems to get ignored:


It isn't about the end result. It is about how you get to the end result. Better data is easier to process. This was evident when I rented the A7s...it was EXTREMELY easy to reduce noise. I very light application of NR sliders in LR was it, even at very high ISOs or with significant shadow pushing. In contrast...the 5D III required extensive NR, after which the data was still not nearly as usable. Additional means would have been required to make the 5D III images as viable as the A7s'...say HDR.


That's my point. Statistically, the NX1 has lower noise, which leads to richer color and contrast (hence the reason the parts of the NX1 image that are supposed to be dark look dark! :P) Better data is better data...and better data is easier to process. For me, it's not just about how B (the end result) looks. It's very much about how I get from point A (the OOC RAW) to point B. Canon doesn't offer me the best data anymore...maybe they never did.

8
All I can say is...WOW. I'm hooked on the NX1!! :P I think it may be my new high speed birding camera some time next year, assuming the lenses pan out. I think I'd get this before I got an A7r even...as I am well and truly impressed.

...an APS-C with BETTER IQ than an FF?? Noo! Say it ain't so! :P

I think Sony has a real competitor. Samsung is another electronics powerhouse...I am very curious to see how they do in the marketplace. I love that they have produced a DSLR-style mirrorless body, instead of some micro-cramped body like the Sony A7 series. I think it is much more along the lines of what I have been looking for, and in comparison to the 7D II...it seems the NX1 tops it in just about every category except lens selection (and, if the adapters work well, that may not even be an issue...and certainly won't be an issue for terribly long, as Samsung is already working on some nice big white supertelephoto lenses.)

Are you the same jrista who so vehemently crossed swords with me a few months ago?


I'm not sure I remember ever "vehemently" crossing swords with you. I think I remember some inane debate over batteries...honestly not sure how the conversation ended up there. But batteries is beside the point. (I also remember you being exceptionally obscure in a lot of your posts...I think people were...befuddled a bit, by a lot of your replies in some of the mirrorless vs. DSLR debates that have occurred in the past. If anyone was "vehement" with you...I'd say it was Neuro...)


Specifically regarding the statement you bolded, I quote myself, from "months ago":


This also causes a drastic loss of battery life ...
Which is solved by bigger batteries ... which is why I keep shouting for people to stop harping on "the small size of mirrorless" as a feature. Make a mirrorless camera as big as 5DIII and cram the sucker full of batteries.
... and the resulting information overload is distracting.  I turn it all of in my EVF cameras ...
For some. But isn't it great that you can actually turn it off, huh?  ;)
... the EVF is lousy in every way compared to an OVF.
Depends. Definitely so in 2012; it became better in 2013; and next year it'll be even better. For comparison, I remember a time when we all felt that film was still soooo much superior to digital and "pros" wouldn't touch it for serious work. But look at where we are today. So please don't judge EVF's on how they are now, as the technology is constantly being improved.  :)
I keep saying similar things (IE use an slr body type, with an EF mount - so no one has to use silly lens adaptors or wait while each and every lens ever made gets resigned to fit the current mirrorless mold).  But, this is where mirrorless has its downfall, it seems like the biggest proponents for mirrorless want their cake and want to eat it too.  All the bells and whistles of an slr, in a package smaller than the A7, but smaller with smaller lenses and of course, a magical battery compartment that can fit 2 1dx batteries...when it comes down to it...it really is about form factor.  there was another post somewhere here showing the first digital camera's, goofy looking things, the first idea was that cause it's new it should look radically different...high tech...result, they looked like a joke and weren't taken seriously until digital camera's started to look like regular cameras.  which is why i feel that mirrorless may just be a cool for now, trendy product.
Aye. Totally agree. I think once mirrorless cameras start coming in DSLR packages, then they will really take off.


If you are claiming I disagreed vehemently with you on the notion that mirrorless cameras in DSLR-sized bodies would be the turning point for them...well, see above. :P I've always believed that...hence the reason I am fairly excited about the NX1. (I still have to see the EVF, it sounds better than Sony's (which definitely have their problems)...I really don't like EVFs, but, if I want the NX1's features...well....yeah...bleh...)

9
Hmm, I am not sure what your using to process, but you are getting radically different results than I am. This is from ACR in PSCC...I simply opened both ISO 6400 NR0 RAW files without any edits in ACR, downsampled the NX1 to the 7D II dimensions, and aligned the images as layers, then cropped the following area:





There appears to be visibly less noise in the NX1 image. The color also appears to be richer, less washed out. The differences in noise can be most clearly seen in the fiddler himself, in the glass of all the bottles, and in the black border and several swatches of the color checker card. The primary detractor that I notice with the NX1 image is the darn CA.


To see if my feelings about the images were correct, I saved the cropped areas as 32-bit TIFF and ran both through PixInsight's statistics tool. I created previews around four areas for statistical testing, and propagated those previews to the other image (so exactly identical regions of each image were compared):





According to PixInsight, the noise on the black part of the Fiddlers Elbow bottle, just under the label, has a standard deviation of:


7DII:  0.048 (R), 0.048 (G), 0.051 (B)
NX1: 0.037 (R), 0.037 (G), 0.038 (B)


Statistically, the NX1 DOES have lower noise. I also checked the maximums:


7DII: 0.325 (R), 0.325 (G), 0.349 (B)
NX1: 0.278 (R), 0.294 (G), 0.302 (B)


The NX1 also has lower maximums, so the black glass there is indeed deeper and richer, with less noise. The preview on the right-hand bottle had the following standard deviation:


7DII: 0.087 (R), 0.061 (G), 0.042 (B)
NX1: 0.077 (R), 0.055 (G), 0.039 (B)


And on the left-hand bottle:


7DII: 0.046 (R), 0.042 (G), 0.039 (B)
NX1: 0.037 (R), 0.034 (G), 0.033 (B)


In every preview, the noise levels of the normalized NX1 image are lower, by a pretty decent margin (enough for the difference to be detected visually.) I also checked the noise levels in the preview around the red part of the fiddlers bottle label:


7DII: 0.067 (R), 0.045 (G), 0.050 (B)
NX1: 0.057 (R), 0.044 (G), 0.043 (B)


Not as much difference in the green and blue channels, big difference in the red channel. That explains the loss in color fidelity with the 7D II...again, more noise, higher standard deviation, so some of the pixels are reaching a brighter/lighter (and therefor, according to color theory, less saturated) "red" tone.


It should also be noted, for maximum clarity here, that the 7D II is at a slight 'advantage.' I downsampled the NX1 image as a whole directly to the same image dimensions as the 7D II image. That did not, however, normalize the objects within the image. It can be clearly seen in my GIF that the NX1 objects are a little larger. This is probably the result of a framing discrepancy. Technically speaking, for a properly normalized test, I should make the objects the same size. If I did so, that would be downsampling the NX1 image even more, thus reducing it's noise even more in comparison to the 7D II.


It should also be noted that I could not find an actual 5D III raw file for download from IR. I looked around, and I could be missing it...but all I could find was a JPEG converted from RAW. That appeared to have considerable color noise in it, so I opted not to even bother using that in this comparison, as newer versions of LR seem to handle Canon noise a lot better than in the past.

10
Third Party Manufacturers / Re: Samsung NX-1 Review
« on: November 26, 2014, 09:07:47 PM »
Canon had difficulty actuating a mirror at 14fps, hence the reason they had to invest special resources to achieve that in the 1D X. But it is no lesser feat to process all that data and achieve the kind of per-frame intelligent processing it does to identify, focus on and track subjects across the frame at 14fps.

The 1DX doesn't actually move the mirror at 14fps, only at 12fps.  In 14fps mode the mirror stays up.


Oh, you are correct. It was the shutter I was talking about for 14fps...sorry, my bad.

11
Third Party Manufacturers / Re: Samsung NX-1 Review
« on: November 26, 2014, 08:07:41 PM »
10 FPS OVF cameras are vastly more complicated, difficult and expensive to produce than EVF mirror-less viewfinders. I'm not sure you appreciate the enormous technical challenge of building a 10FPS reflex system. These camera are engineering marvels and only a few companies in the world have the capability to build them.

The EOS-1v of ca.2000 vintage could do 10fps ... so this feat is just plain old hat today.

If you are interested and not just trolling...

The tricky part is not making the mirror go up and down ten times a second, it's all of the things that have to happen between each frame while the mirror is locked down for a thin fraction of a second. Focus tracking, evaluative metering,  iTR face recognition and vibration damping and a whole host of other things which are still done off of the mirror. In addition the mirror has to be down long enough to allow the photographer to continue viewing the subject and compose the frame during fast paced action while all of this is happening. Then the mirror has to  get out of the way of the shutter but without causing any image degrading vibration.

If you had any appreciation for the engineering feat this represents I doubt if you'd still go around calling them mirror-slappers. But maybe you would, who knows.


Aye, there is a lot more going on today than back in the EOS 1v days. Back then, you read data from a rather simplistic (by today's standards) AF unit, flipped up the mirror and actuated the shutter, then forwarded the film.


Today, we have to read millions of pixels off the sensor, on top of everything else GraFax described. The AF units of today are vastly more complex in design, using higher density AF points with more pixel-rich phase-detection stripes. AF and metering systems are integrated, allowing things like iTR, all of which requires fairly significant horsepower to drive...all at 10fps.


Canon had difficulty actuating a mirror at 14fps, hence the reason they had to invest special resources to achieve that in the 1D X. But it is no lesser feat to process all that data and achieve the kind of per-frame intelligent processing it does to identify, focus on and track subjects across the frame at 14fps.

12
Some more information on this:


http://image-sensors-world.blogspot.com/2014/11/active-pixel-color-sampling-story-goes.html


The commentators still seem to think it's a hoax, given the discrepancies and errors in all of the data. Take it for what you will.

13

Given the kind of madhouse progress being made in the sensor world right now, I totally believe it. Especially since Red already produced a 21 stop sensor some time back, which in practical use gives up to around 18 stops or so of real-world DR.

to get 21 stops, you would need 2^21 or 2,097,152 electrons in the well and A/D circuitry good enough to count each electron.... and that means more than that number of photons.... unless we are talking LARGE pixels, it seems a bit unlikely to me...

The problem with traditional sensors and measuring DR is that is based around powers of 2. This means we have a lot of values made for highlights that are traditionally wasted.

I'd love to see someone come up with a different design that allocated sensor values evenly across the DR space or at least made it possible to have as much value space for shadows as for highlights.


This is a common fallacy. Sensors are linear, in that they accumulate electrons in a primarily linear fashion until you ride up to the clipping point (where you MIGHT experience some slight non-linearity). There is no "wasting" values for highlights. Pixels simply accumulate light over time, there is no curve.


The notion that the dynamic range of a sensor is wasted on highlights is a mistaken one derived from the fact that RAW editors apply tone curves when rendering the bayered data to make it SEEM like that is the case. It's an illusion, though. The tone curve is applied after the data is captured off the sensor...the sensor is still and always will behave linearly.

14
Canon General / Re: Does Canon really deserve this?
« on: November 26, 2014, 06:52:42 PM »
As far as I know, Canon is the leading camera maker in the world, so they are doing something right.

However, as with cars, there is a lot of competition and the differences between makers is not relatively great. In such a market, it's easy to quibble on various points, and make mountains out of mole hills. Not that all detractors are doing so. There are big things like the inferior dynamic range of Canon sensors when compared to Sony/Nikon cameras. Then there is the lack of competitive mirrorless lineup. However, the dual pixel sensor technology is way ahead of everyone else, and gives Canon a big advantage in the video market and will serve them well when they do produce a viable mirrorless product. So they are far from out of the game, but they are being pushed, and that is a good thing for us, the photographers who use their products.


It's only a good thing if Canon actually steps up to the plate they are being pushed towards. It should also be noted that DPAF-like technology (and some of it much more advanced) is already being patented by Canon's competitors...so the edge will not remain Canon's for long.

15

Given the kind of madhouse progress being made in the sensor world right now, I totally believe it. Especially since Red already produced a 21 stop sensor some time back, which in practical use gives up to around 18 stops or so of real-world DR.

to get 21 stops, you would need 2^21 or 2,097,152 electrons in the well and A/D circuitry good enough to count each electron.... and that means more than that number of photons.... unless we are talking LARGE pixels, it seems a bit unlikely to me...


Or multibucket technology in each pixel. Stuff a bunch of CCDs for overflow capacity in some largish pixels, and I could see it happening. I suspect there is some king of fudging going on though, like Red Dragon (they claim 21 stops as well, but in reality it seems to be "18+", which is around a 300ke- FWC, which is certainly more doable.)


BTW, there is EMCCD technology, which effectively IS a photon-counting CCD sensor. They use electron multiplication to super-overwhelm read noise, effectively negating it entirely. Now, EMCCD has to be cooled to insane temperatures (I think a minimum of -70°C, since any stray electron caused by dark current will be multiplied and counted like any photon), so that certainly limits it's applicability.

Pages: [1] 2 3 ... 312