December 11, 2017, 10:46:15 AM

Author Topic: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision  (Read 4060 times)

aceflibble

  • EOS M5
  • ****
  • Posts: 171
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #15 on: December 08, 2017, 09:41:40 AM »
I also don't think the processor makes any difference at all to the RAW images at all. The processor and jpg processing algorithms do make a difference to the quality of JPG images, but those are not what is being compared in the video. (RAW images are being tested in the video)
'Raw' is processed too.

Basically, all the sensor does is collect the light; it's the processor which interprets it.

Yes this applies to 'raw'. No, it's not just for .jpgs. There is a huge, huge difference between the processor within a camera and a 'processor' as in software converting to a different format. They may be frequently called the same thing but they are entirely different.

'Raw' files are not actually truly 'raw' data. To create the 'raw' image file the processor needs to interpret the light collected by the sensor.

The way I've found people easily understand this in the past is to liken it to food production, so bear with me a second.
Think about basic food ingredients such as flour or sugar, which you might use to make a cake. When you buy them they seem to be in a pretty basic form—'raw', you could say—but you know that before you were able to buy that flour and sugar it first had to go through many stages of production. It had to grow, be harvested, cleaned up, processed and packaged up. Only after the ingredients have already gone through all that can you buy them and then turn them into a cake.

You can think of the camera's processor—remember, that's the psychical processor driving everything, not the software—as being that intermediate stage between food first being grown and you getting to cook with it. When you open your 'raw' image and start to 'process' it, that's the equivalent of you starting to bake that cake; the base ingredients you're working with were already harvested (sensor) and cleaned and packaged up (processor).

Another way to think of it is like a solar panel. The panel—that's the camera sensor—can collect the light, but it requires much more—the processor—to actually turn that sunlight into energy, which then can be used to power whatever you want. (That last bit is your post-processing.)

 
So, what is the camera's processor responsible for? It's taking the light the sensor reports and puts it all in order. Noise is mostly down to the processor as it tries to interpret variations in the under-stimulated sensor. The processor decides what data can be safely discarded (yes, even with lossless 'raw', some data is lost) when multiple neighbouring pixels report exactly the same information. (This is the basis of what we commonly refer to as dynamic range). When you see colour banding or specific noise patterns, that's usually down to the processor.

If you look at Fuji cameras, you can see how important the processor is compared to the sensor. Their 'A' cameras use a bayer sensor while their other 'X' cameras us a unique 'X-Trans' array. Despite the sensor's pixels being in a different order, though, all their cameras end up with the same look to the raw files because their processors are the same. A lot of people think DxO don't measure Fuji cameras because of the X-Trans array, but DxO can't measure the X-A bayer camera either... because it's really the Fuji processor that is getting in the way.

This is also why Nikon and Sony cameras have varied in image despite them using the same sensors. They use the same sensor, but not the same processor. As a result, the basic interpretation of colour, contrast, brightness, and noise are different.

Canon users of a decade or so ago, especially portrait photographers, will remember how big a deal it was when DIGIC II came along, and then again when DIGIC II was replaced, and how colours and contrast in Canon cameras changed so much in that period even though the sensors themselves mostly hadn't changed. (Especially the APS-C sensors). There are still some people who swear by DIGIC II cameras (and the first generation of Fuji processors, for that matter) as producing the best files for portraiture.
 

This is all done to the 'raw'. You are never getting truly untouched 'raw' data. If you did you wouldn't be able to open it. Every 'raw' file, from every manufacturer, has to have been processed by the camera's processor—again, this is the physical chip we're talking about, not the software .jpg conversion—before you can open it.

And before anybody asks "well why don't they give me the option to have the file unprocessed, and why doesn't someone develop a way to open that truly raw data?" the answer is quite simple: imagine buying a lens which never allowed you to focus it, ever. Yeah, now you see why you don't want truly raw data.

 
Final word on the subject: it would really help ease confusion if people got used to being more specific with their terminology. When you're talking about editing a raw file be sure to call it "post-processing", not just "processing". Don't call Photoshop a "processor", for example, but a "raw processor" or "post-processor" are fine.
There can be up to five different stages of 'processing'from the time the shutter is pressed to the time the file is finished and you typically use at least two physical processors in that process, so you can see how the process of using the correct names for each processor used in the process will make the process of discussing processes and processors less confusing.

Yeah, try and get your heads around that one.
 

You are splitting hairs - it is whether the D850 as a camera is better than the 5DIV as a camera. People use 'sensor' as a shorthand because it is the easiest thing to talk about even though the sensor provides the raw material. Yes, a manufacturer could screw up a great sensor with a poor processor but what is really important is whether you can get good photos from it.
When you want to get an idea of what these parts can do, so you can start to get an idea of what future cameras may be like, it's very important to make the distinction.

E.G.
The 7D3 may well use the same processor—in fact it will likely have two of them—but a different sensor. Knowing which part is responsible for which aspects of the 5D4's image quality can help us estimate the nature of the 7D3's image quality, which helps inform early adopters. Conversely, if you ascribe everything simply to the sensor, you've learnt nothing about the next camera.

canon rumors FORUM

Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #15 on: December 08, 2017, 09:41:40 AM »

Geek

  • PowerShot G1 X II
  • ***
  • Posts: 27
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #16 on: December 08, 2017, 12:48:25 PM »
Basically, all the sensor does is collect the light; it's the processor which interprets it.

Yes this applies to 'raw'. No, it's not just for .jpgs. There is a huge, huge difference between the processor within a camera and a 'processor' as in software converting to a different format. They may be frequently called the same thing but they are entirely different.

'Raw' files are not actually truly 'raw' data. To create the 'raw' image file the processor needs to interpret the light collected by the sensor.

I think you are missing part of the equation here.  It is the Analog to Digital (A/D) converters that are one of the more critical parts in the path of getting a picture from the image sensor to a usable raw file or jpeg image. 

The A/D converters and associated amplifiers and other analog signal conditioning components are responsible for converting the analog data collected by the sensor array into the binary data that the processor(s) can handle.

Until recently this has been a big differentiator between Canon and Sony/Nikon camera systems.  Sony sensors and A/D converters have been manufactured onto the same piece of silicon giving them an advantage in signal noise and therefore cleaner final images.  Canon has typically used A/D converters that are on separate pieces of silicon and require longer paths and more potential for noise in the analog path before getting to the A/D converters. (Read some of the other threads that cover this in far more detail than I'll bore you with)

The processors (Canon Digic - Digic 7, Sony BIONZ X, Nikon whatever) take the digital data from the A/D converters and process it.  Some if not all of the processors have extra Digital Signal Processor (DSP) hardware in addition to the general purpose cores.  The DSP's are used to accelerate processing the digital data in the processors.

The reality is that all processors only process the digital data that is given to them from the A/D converters.  The difference between the original Digic and the latest Digic 7, 8 or whatever is the performance of the processor and the power usage.  With the appropriate firmware, an original Digic can do the same thing that the latest Digic 7 can do, only much slower and using more power.  The Canon Digic  line uses an ARM core (a general purpose processor) and some specialized DSP hardware to accelerate processing the digital data.

The processor version make no difference on the final image, only the sensor, the A/D conversion and the algorithms used to process the digital data.

And yes, the raw file is the digital data from the output of the A/D converters along with some other information about the required to turn the data into a usable image.  I'm sure as mentioned above there is some processing applied to the raw data to reduce noise adjust for ISO levels, etc. But that is relatively minimal compare to the processing required for final images.

So in summary, the sensor and A/D conversion are very critical to the images.  All of the work done by the camera to insure proper exposure and focus is critical to the images.  Even the image manipulation algorithms are critical to the images.  However as long as the processor is fast enough and does not use too much power, it really does not matter which one is used or even how many are used.
EOS 7D mark II, EOS 40D, EF 16-35 f4L, EF 24-105 f4L IS, EF 70-200 f2.8L IS II, EF 100-400L II, EF-S 18-135 STM, EF 28-135, EF 70-300, 580EX II

stevelee

  • EOS M5
  • ****
  • Posts: 224
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #17 on: December 09, 2017, 01:58:27 PM »
I don’t guess any usual camera gives out the raw data from the little sensors under the green, red, and blue filters.

neuroanatomist

  • CR GEEK
  • ************
  • Posts: 21810
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #18 on: December 09, 2017, 03:57:18 PM »
I don’t guess any usual camera gives out the raw data from the little sensors under the green, red, and blue filters.

Those data are in the RAW file, the demosaicing is performed by the RAW converter (in-camera or on your computer).  Typical converters do not allow you to access those data, you need more esoteric software (e.g. RawDigger, Rawnalyze).
EOS 1D X, EOS M2, lots of lenses
______________________________
Flickr | TDP Profile/Gear List

9VIII

  • EOS-1D X Mark II
  • *******
  • Posts: 1675
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #19 on: December 09, 2017, 11:34:56 PM »
...
And yes, the raw file is the digital data from the output of the A/D converters along with some other information about the required to turn the data into a usable image.  I'm sure as mentioned above there is some processing applied to the raw data to reduce noise adjust for ISO levels, etc. But that is relatively minimal compare to the processing required for final images.
...

This paragraph is self confliciting.
All RAW files are baked, it is impossible to access data as it was generated directly from the ADC on any modern camera. The design of the processor will affect the final image. Hook up Digic 4 to a 5DS and you’ll get a very different image, not just get it slower.
Maybe at one point things weren’t processed that way but they are now.

aceflibble

  • EOS M5
  • ****
  • Posts: 171
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #20 on: December 10, 2017, 05:38:41 PM »
[words cut for space]
You're overstating how much the ADC is doing and also how distinct it is from the rest of the processor. The ADC matters, yes, and in past generations some of the biggest leaps in image quality and possibilities were down to advancements in ADC efficiency, but for several hardware generations now we've been at a point where the ADC is A) capable of far more than is being asked of it, and B) so refined and 'direct' in its work that it really can't have any particular image qualities attributed to it. Additionally, it has been a long time since it was normal for the A-D conversion to be handled by a physically separate unit than the main processor (I say 'main'; of course many SLRs have been made with multiple CPUs, and which one is doing the most work varies from camera to camera) and in many cases the whole imaging sequence—light capture by the sensor, conversion to digital, and processing and saving—is now handled within what is technically a single part (by patent purposes), rendering the distinction between ADC and the CPU not only irrelevant but actually inaccurate.

Analogue-digital conversion matters, but your take on it is a little out of date and even in the cases where it's not (e.g. Fuji cameras) it's still not really correctly crediting each part.

Don Haines

  • Canon EF 600mm f/4L IS II
  • **********
  • Posts: 6294
  • posting cat pictures on the internet since 1986
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #21 on: December 10, 2017, 06:19:44 PM »
Old tech.... the 7D2....

By going to an external A/D system like the traditional Canon system, you have 8 (or 16) A/D converters, and 20 million pixels to read at a burst rate of 10FPS.... That means that in a dual DIGIC setup, the A/D has to be able to read a new value every 80 nanoseconds.... to read video at 60FPS, you are now looking at a new reading from each A/D every 13 nanoseconds... Darn Fast! This is why there is no 4K on the 7D2.

Then switch to new tech.....

Compare this to to an on-chip A/D.... you have one for each row.... for the same size sensor you are now looking at 5472 readings times 10 frames per second..... or 18,265 microseconds per reading..... a heck of a lot slower, and that means greater accuracy and much lower noise.

The change to the type of A/D system means far more than which processor is running the algorithm to place the output data into a file......
The best camera is the one in your hands

canon rumors FORUM

Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #21 on: December 10, 2017, 06:19:44 PM »

Spock

  • PowerShot G1 X II
  • ***
  • Posts: 29
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #22 on: December 10, 2017, 06:56:44 PM »
...
And yes, the raw file is the digital data from the output of the A/D converters along with some other information about the required to turn the data into a usable image.  I'm sure as mentioned above there is some processing applied to the raw data to reduce noise adjust for ISO levels, etc. But that is relatively minimal compare to the processing required for final images.
...

This paragraph is self confliciting.
All RAW files are baked, it is impossible to access data as it was generated directly from the ADC on any modern camera. The design of the processor will affect the final image. Hook up Digic 4 to a 5DS and you’ll get a very different image, not just get it slower.
Maybe at one point things weren’t processed that way but they are now.

read settings from camera and lens, store in file in the proper data format and location as per file spec definitions.....

read values from output of A/D converter in proscribed pattern, use predefined algorithm for lossless compression, and place in file as per file spec definitions...

save file.

It is the file definition and the computing algorithm that are important. If done properly, you only get one answer.

The processor does not affect the results. this is a lot like saying that the value of 16 bit integer 6 multiplied by 16 bit integer 9 depends on what computer you use to calculate it with...

RGF

  • Canon EF 300mm f/2.8L IS II
  • ********
  • Posts: 2442
  • How you relate to the issue, is the issue.
Re: Dustin Abbott Nikon D850 vs Canon 5D mark IV sensor comparision
« Reply #23 on: December 10, 2017, 08:38:27 PM »
Thought this was very good.  I am surprised how close the 5D M4 comes to the D850.

midluk

  • EOS 80D
  • ****
  • Posts: 278
Even in cases where the ADC is embedded in the same chip as the (digital) processor, I would not count it as part of the processor for this discussion. For me the ADC is part of the sensor (even if not on the same chip). The ADC of course has an influence on the noise performance of the image, but not on color (except for cases where it introduces crosstalk between pixels of different colors).

I have played around with the raw files (self-made c++ program using libraw) from the 70D and there are at least two areas where it is appears to not be completely raw. The raw file contains meta data (including the preview-jpeg) and a 14 bit number for each pixel. Each pixel is sensitive to only one color. On the edges of the sensor there are some special covered pixels, some black, some always saturated.

The black level in the file is always 4096, independent of exposure time and ISO. I would have expected that to go up due to dark current. It seems like the camera is subtracting (or adding) an offset to all values to shift black to 4096. The camera can compute the offset from dedicated pixels at the edge of the sensor which are covered by some material and therefore always receive no light. It is however not completely impossible to do the offset completely analog before the ADC, so the processor might not actually alter the data from the ADC here.

Known bad pixels (a list of bad pixels in the camera) are masked (likely replaced by the average of the surounding pixels). My camera has developed two or three hot pixels during the years, which no longer appear in the raw file after letting the camera check for them after or during manual sensor cleaning (close body cap, set camera into manual sensor cleaning, wait ~30s, turn camera off, wait some more). The fact that they are not removed automatically before making them known to the camera leads to the conclusion, that no additional noise removal (like Sony's star eater) is performed on the raw data.

The processor can do nothing for color (except for adding meta data and the preview jpeg) in raw files. If changes in color appear with changes of the processor, this might be caused by changed color filter arrays which produce better results but need more processing power to generate the in-camera jpeg (which only the new processor can do reliably). Or just changes to the default processing profiles in the camera which then are also included as default in raw converters.
5D4, 70D+BG E14, EF 24-70 f/4L IS, EF 16-35 f/4L IS, EF-S 17-55, EF-S 10-22, EF 100 f/2.8L IS macro, EF 70-200 f/4L IS, EF 100-400 L IS II, 2x 600EX-RT, 430EX III-RT

sanj

  • Canon EF 300mm f/2.8L IS II
  • ********
  • Posts: 2916
Thought this was very good.  I am surprised how close the 5D M4 comes to the D850.

Mean.

canon rumors FORUM