Hi Dave, Great video! However your analysis method has some fundamental errors. You did not take pixel fill factor into account. A basic CMOS pixel is usually made up of 3 to 5 transistors (MOSFETS in this case) as well as a photo-diode (the light sensitive part). All of these devices must fit in one pixel. The ratio of the area of the entire pixel to the area of the photo diode is called the fill factor and it is a far more important metric than pixel size to consider when evaluating noise performance. A 10μm x 10μm pixel with a fill factor of 25% has the same photo-diode area as a 5μm x 5μm pixel with a 100% fill factor. At a given photon flux (#photons/area*unit time), a larger photo-diode surface area will result in more photons being absorbed per pixel. You can have many different sized pixels that all have the same fill factor depending on there architecture. One reason Sony sensors are so popular among camera manufacturers is because of their Backside illumination architecture. They are able to put the transistors UNDER the photo-diode in each pixel, this results in very high fill factors approaching 100%. The specific pixel architecture also needs to be taken into account. For example, if you want really low noise you want to amplify the signal as much as possible within the pixel itself. This could require more transistors which take up more space. Some sensors even have ADCs and memory under the pixel itself to minimize analog noise during readout!! If you want a global shutter you need lots of extra transistors. If you want good dynamic range you want a large electron-well capacity -> bigger photo-diodes...... trade-offs, trade-offs, trade-offs. When a photon of sufficient energy ( > 1.12eV for Si ) hits the photo-diode it generates an electron-hole pair (excess carriers). Basically this means there is now a free electron and hole (absence of an electron) to move around and generate a current. By applying a reverse bias voltage to the photo-diode, we can measure changes in the resulting current due to changes in the excess carrier concentration. The excess carrier concentration is a function of the light hitting the diode.... and sadly, as you mentioned, heat. Thermal energy is the unwanted source of excess carriers in the case of a image sensor. A large part of the image noise we see is due to thermally generated electron-hole pairs and the inherent leakage current of the diode. This brings me to another point on why sensor/pixel size cannot be directly compared. Each sensor could be made out of slightly different material!! Most image sensors today are made out of silicon, this silicon is "doped" with other elements to alter is electrical properties. Different types and concentrations of dopants lead to VERY different properties. Changing the dopants changes how the photo-diode and all semiconductor devices will operate and respond to light. For example, you could have two silicon CMOS sensors, both full frame , both 24MP, both with 90% fill factor and same readout circuitry / ADC .....if the Si is doped differently they could have different noise performance. Micro-lenses are another thing to take into account. Some sensor architectures have little lenses above each pixel that concentrate the light on to the photo-diode area of the pixel, this increases the effective fill factor. (like a Metabones speed booster but at the pixel level!) My point is, engineering is all about trade-offs. Nothing comes for free. You claim: "PIXEL SIZE doesn't affect noise performance" .....but actually it does, due to the nature of design trade-offs, very much so :)
Modern sensors are BSI, with microlenses over every pixel, so fill factor is kind of irrelevant. All sensors are good at it and perform roughly the same. They are also made by Sony, mostly using the same tech. (this is very apparent on the dpreview studio shot comparison)
@@93kristof Yes Sony makes a large number of sensors but not all. Fill factor is still an important metric as there are many architectures and implementations. Not all modern sensors are BSI, Sony them self still make front-side illuminated products ( IMX323LQN, IMX225LQR etc) . All modern sensors do not preform roughly the same ...for example compare a CMV12000 , CMV50000, a STARVIS and a Fairchild HWK1910A. All rather modern devices with totally different performance.
@@Shane_McLachlan We are talking about mainstream camera sensors here that you can buy. Those all perform roughly the same, except Canon, which is like half a stop behind due to their ancient tech. Not even BSI makes much of a difference because of microlenses. So yeah, for actual consumers buying cameras that exist - fill rate, pixel size and what not - none of these matter, just the size of the sensor.
@@93kristof You can buy sensors like those in mainstream cameras. Blackmagic Design uses Fairchild and CMOSIS (now AMS) sensors in alot of their products. For example I have a CMV12000 in a Production 4K (also used in URSA Mini 4k). It's a very interesting sensor, uses 5T pixels with global shutter readout (hench the FPN ugh). Sure, maybe all this stuff doesn't matter to consumers... That doesn't mean it's not true ;) I would love to see digital filmmakers/photographers educate themselves more about the inner workings of digital cameras systems. Back in the film days that was far easier cause the concepts were easier to grab. Currently alot of filmmakers/photographers operate on a high level of abstraction as to what actually goes on in a digital camera. This leads to misconceptions.
@@Shane_McLachlan I would say the opposite, people are too much interested in gear. At least if RU-vid content and comments are to be believed. This detail stuff about micro lenses etc. is completely useless information from the perspective of taking good images. It is enough to know what are the ruff differences between cameras, their specs, features and performance, not the technical details behind them.
While down sampling images will greatly reduced noise, the pixel binning is performed after voltage quantization. This means that any inherent photon shot noise will be averaged out but will not give the same performance as pixel averaging in camera. It can come close if the camera electronics are very noise free, however larger pixels, or CCD chips that can bin before quantization will always show less noise given the same noise levels recorded.
I came across this video after watching your does larger sensor give u better exposure. Glad i watch these 2 video. Since most internet claims u need a full frame camera for better pics. I'm glad i didn't trashed my m50 for a full frame
The sensor size is the only thing that matter in terms of noise BUT it's not due to the spread out circuitry (It's not the main reason). Per surface unit, all sensors are producing about the same noise, BUT the resulting picture will be displayed at the same size whatever the sensor size is! Increasing the picture size to view all at the same size increase the noise. The same way if you crop your FF picture to APSC size, you'll have the same noise as your APSC camera not croped. As well if you look at your uncroped APSC picture 0,67x the size of your uncroped FF picture.
The problem with your analysis here is that when you down sample you get a sharper image with more dynamic range and less noise, but that isn't what I want. I want a soft image to use with some softer lenses, not crazy sharp, but more like a retro style image like the original Charlie and the Chocolate Factory. You can use something like a 7Artisans APS-C cinema lens on the ZV-E1, A7S II, FX3 or FX6 and clear image zoom in 1.5x, and you're up sampling the image. For this pixel size matters.
So Dave, you are telling me that pixel density doesn't matter... Why don't full frame camera manufacturers already have sensors with 80MP, 100MP, 120MP etc.? A 20MP micro four third sensor is roughly a quarter the area of its full frame brethren thus it should be manageable to produce an 80MP full frame camera, with little to no impact on noise and dynamic range, according to your theory. Pixel density does affect noise performance, just not to the extent it once did, as a result of continuing improvements in software and processing power. This transition away from pixel density restrictions actually bodes well for APSC and MFT sensors!
@@DaveMcKeegan Cost and likely performance, was not appealing enough in justify the resolution increase for the main steam (more likely a sensor designed for a specific application)... That still does not explain or answer why full frame format camera options are not currently available at 80+ MP, considering that an equivalent pixel density is found in MFT cameras. If it was just the total size of the sensor that mattered for sensor performance it would be a no brainer for a camera manufacturer to produce an FF 80MP camera appealing to landscape photographers. The reason they have not is a result of the top of the line/latest sensor technology not being able to maintain the expected full frame quality of lower resolution cameras within this format. Yes, they could easily produce an 80MP sensor today but it would come with the trade off of performing similar to a MFT sensor, which would defeat the purpose of increasing the cameras resolution. Everything has trade offs.
Because as well as increased cost, you have to allow greater processing power in the camera to handle it, larger file sizes which slow down work flow ... And realistically at the minute there isn't enough demand for such high resolution sensors to justify it ... But they will in the future. A few years ago 40-50MP seemed impossible and now it's normal, Sony G-Master lenses are designed to render up to 100MP so clearly Sony plan those sorts of resolutions in the future
@@DaveMcKeegan Cameras have been processing 80 MP files for over 2 years now since pixel shift technology was released. Processing is clearly not the issue. Obviously this type of cameras would not be marketed to sports and wildlife enthusiasts (too much processing for fast CAF shooting). Fujifilm is releasing a Medium Format camera next year that will have a 100 MP sensor, yes its expensive but still being released because they think it would sell. A full frame manufacturer could easily under cut the cost if they decided to make one them selves, so why don't they? Its because at the moment, sensor technology is not good enough to maintain the same image quality as their lower resolution cameras, as a result of the increase in pixel density. A FF 80 MP camera would be reasonably affordable within the high end camera market of today because in essence it would be the equivalent of 4 MFT sensors which are not overly expensive and cameras have been processing 80 MP photos for over 2 years.
As already mentioned, Canon released a working 250MP (smaller than full frame sensor) with sample pictures so the technology already exists. As for pixel shift, firstly you don't get any extra pixels in the finished image, the image remains the same physical resolution but you get more detail because it combines red, blue and green pixels together. Also (at least in the A7RIII) the camera can't process the finished file, it simply takes 4 regular images and you merge them in computer later
I was thinking about this the other day and I hadn't put all of my thoughts together yet. Thank you for doing it for me Dave! Now I don't have to think for myself! 😁
I think those pixel sizes are incorrect. I've only seems scientific grade sensors with pixel size greater than 10 μm. In a quick search I found the sensors used on a couple of sony cameras: - A7III uses IMX251 with a pixel size of 4.50 μm; - A7IIIS uses IMX510 with a pixel size of 3.76 μm, but it does 2x2 binning, merging 4 sensor pixels of 3.76 μm x 3.76 μm in one image pixel. The cooling is used in some astro cameras is to reduce dark current on long exposure. Ironically, astro cameras with smaller sensor often don't have cooling because their use case.
My mistake, I'd quoted the pixel area value instead They are still linked as it's the area each pixel covers where as pixel size measures the distance from the center of one pixel to the centre of the next
Dave, I just stumbled across your great video. While your conclusions are spot on, the reasoning is a bit shaky (regarding scaling down) and IMHO even wrong (regarding the better lower noise with bigger sensors). Both might lead to some of the commenters simply denying your findings („higher iso creating noise I'm not dumb … " blahblah) By chance, I had the same problem: To convince people that noise is a matter of sensor size, not pixel size, so I wrote down the reasoning lately. I hope you don’t mind me sharing it here as a comment. Sorry, it's a bit lengthly and contains a tiny bit of math ... Noise when scaling down images Scaling down an image comes down to combining multiple adjacent pixels into one. If we scale by the factor of two (MP wise), we - in average - combine two pixel into one. If we think of the content of each pixel as a signal that is polluted with noise, then we need to add the noise in a different way than one would add the pure, non-polluted signal. Let s1 be the amount of pure signal and n1 be the amount noise. Then to combine two adjacent pixels with the same amount of signal and noise (s2 and n2 being the combined amounts) … .. we add the pure signals: s2 = s1 + s1 = 2*s1 … BUT add the SQUARES of the noise: n2 = sqrt(n1^2 + n1^2) = sqrt(2*n1^2) = sqrt(2) * n1 The latter is called „addition in quadrature“ and must be used whenever we add the amount of two signals that are _uncorrelated_ (handwiki.org/wiki/Addition_in_quadrature). The noise in digital images is to a very large part uncorrelated (i.e. random) noise stemming from the physical properties of light (shot noise) and sensor electronics (thermal noise etc.) Why we need to use that „addition in quadrature“ thing is beyond that comment, but it is _purely justified by mathematics_ rather than experience or feeling or visual reception or whatever. Back to our little calculation: The ratio between signal and noise is called „Signal to Noise Ratio“ or SNR. The higher the SNR, the less noise we perceive in an image. The SNR of the unscaled image is SNR1 = s1 / n1. And with the calculation above the SNR of the scaled down image is SNR2 = s2 / n2 = (2 * s1) / (sqrt(2) * n1) = sqrt(2) * (s1 / n1) = sqrt(2) * SNR1 So: SNR2 = 1.4 * SNR1 _What all of this simply means:_ If you scale down an image by factor 2 (e.g. from 48MP to 24MP), the SNR increases by factor 1.4 (in other words: the noise decreases by 30%). Btw, the same mechanics work for other scaling factors. Factor 4: SNR2 = sqrt(4) * SNR1 = 2 * SNR1. Hence the noise decreases by 50%. _This is true regardless of camera or ISO!_ Not believing me? Have a look at DXOMark. To do so, you need to know that SNR in DXOMark is (as usually) expressed in Decibels (dB). A factor of 1.4 is roughly 3 dB, a factor of 2 roughly 6 dB, a factor of 4 12 dB and so on. In a single compact formula: SNR2 = 10*log(a) + SNR1 (with SNR in dB). Now select ANY camera you like and have a look at the measurements (e.g. www.dxomark.com/Cameras/Nikon/D5300---Measurements). Select the „SNR 18%“ diagram. First select „Screen“ and hover over one of the dots. Note the SNR. Here e.g. 36.9 dB at ISO 100. Now select „Print“. Don’t be fooled by the name „print“: This diagram simply shows the SNR of the image scaled down to 8MP (this information can be found in the tooltip of the „Print“ button). The D3500 in the example has a resolution of roughly 24MP (see specification tab). Hence the image in „Print“ was scaled down by factor 24 / 8 = 3. Put into our formula, the SNR of the scaled image should be 10*log(3) + 36.9 dB = 41.7 dB. Hover over the corresponding dot in the „Print“ diagram. Et voila: 41.7 dB at 100 ISO! You can try this with any camera at any ISO. It will be always be true. Sensor size and noise If one has two cameras, a FF and a APS-C and all other variables the same (subject, light, optics, exposure), then the illuminance in LUX, i.e. the amount of light _per area unit_ , that hits the sensor surface is obviously equal. But since a FF sensor has approximately twice the surface area than a APS-C sensor, the _total_ amount of light that hits the FF sensor is approximately twice as much. Hence, if both sensors have the _same pixel count_ (say 24 MP), for the FF sensor each pixel (photo site) has twice the size (area) and consequently collects twice as much light. Due to physics, twice the light translates to a sqrt(2) (or +3dB) better SNR (at least if shot noise dominates, which is more or less true as long as you are not into things like astro photography). In contrast, if both sensors have about the _same pixel size_ (which means the FF sensor has about twice the pixel count, say 48 MP), each pixel (photo site) collects the same amount of light for both sensors. Hence, the SNR for both sensors will be the same. BUT, since you now have twice the amount of pixels, you can scale the resulting image down to 24 MP which - according to the explanation above - gives you an increase of SNR by +3dB. Which is exactly the same as with the FF 24MP sensor! Long story short: Let’s compare three fictional cameras: 1. 24 MP FF 2. 24 MP APS-C 3. 48 MP FF Let everything else, optics and particularly sensor technology and generation be equal. Comparing 1 and 2, 2 gives you finer pixels and hence more room for magnification in post (fine for shooting birds), while 1 gives you a +3dB better SNR. And 3 gives you the best of both worlds, but _not at the same time_ ! You only can _either_ enlarge _or_ enjoy the higher SNR. This is why it’s sometimes said a D850 or a Z7 is „two cameras in one“. Unfortunately this is even true if you look at the price tag …
Without being argumentative, you state that sensor size affects noise more than anything, but then state a half-a-dozen other variables that also affect noise. Testing your theory out, I used the website DPReview.com and checked the image quality of all 4 Sony cameras side-by-side. Unlike the DXOMark's views on ISO scores, the actual photos show a precise and direct correlation of noise performance to pixel size. I agree with your assertions that other factors attribute to noise REDUCTION, but that doesn't mean that the correlation of pixel size isn't the primary cause of noise BEFORE algorithms are introduced to mask it.
Because the DPReview noise tests are shots displayed at 100% which isn't an accurate representation of how we'd actually view the images and is also then subject to the opinions of the viewer DXOMark measures the noise ratio from an 8MP downsample so is a consistent reading You can download the DPReview RAW lab test images, if you do and scale each image to the same size and the view it you would see a different result.
@@DaveMcKeegan I can see a difference. I don't see why you can't. As far as not viewing at 1-to-1 if you view a vista through a window with a screen, and stand far enough away, you might not notice the screen, but the screen still affects the vista, regardless of perception.
Except that there is a little gap around each pixel, so the total area available for light gathering is less for high mega pixel cameras than low. Think of the margins around the text on a page. Larger page, more text per page.
You nailed it on sensor size being the main determinant of noise performance, but you're wrong about electronic noise being the main source. The biggest contributor of noise is actually shot noise, aka photon noise, aka poisson noise. It has to do with the random nature of light (think of it like rain where raindrops don't land uniformly on a surface), the larger surface of a bigger sensor reduces the variance of photons. dpreview goes more in-depth about this: www.dpreview.com/articles/8189925268/what-s-that-noise-shedding-some-light-on-the-sources-of-noise
How does this relate to video? And if it still applies the same way, why is the a7sii considered an incredible low light camera and the a7iii and a7riii aren't? Also do you know it the fx3 has better noise performance than the a7siii due to it having fans?
Youre wrong, matey. Smaller pixels produce proportionally SMALLER digital grain. When I bought my 50 megapixel 5dsr I pretty much immediately thought "so wheres all this extra noise I'm supposed to be seeing". The noise was there but much smaller in size so Id have to zoom right in to see it. At a viewing of the full image it was way too fine to be able to discern. So when I go reseaching databases that measure cameras noise levels, sure enough, I I find my 50 mpx dsr has better measured noise than my 22 mpx 5d mkiii. And thats exactly how it looks.
Great video. Combination of pixel size and circuitry for original noise level. Then down sampling significantly reduces the noise. You can even see the down sampling effect in cameras that down sample video. They have cleaner video. This is why even on my APSC cameras I use ISO up to 32000 for event images and not once has anyone mentioned noise. Because at the resolution they are viewing the image at, there is no perceivable noise. And the more pixels you have on a sensor the more you down sample for any given size image. In fact when down sampling takes place you can get smoother gradations in colour as it averages out the noise. When image editing you can produce smoother transitions by introducing noise and then denoising the image.
Downsampling helps with noise ofc, but you also see the benefits of it in noise-less videos (very good lighting etc), and the reason for that isn't because downsampling reduces noise. A big reason downsampling helps video is because sensors use a bayer filter, which means you do NOT have a full RGB sample per pixel; the pixels are interpolated with several algorithms to fill the missing information between the pixels. Let's say you're shooting 1080p video by cropping a common digital camera sensor; the resulting information is NOT 1920*1080*3 samples per frame; you'll get much less than that, and so you it will be impossible to get a "fully sharp" video doing that. Now if you downsample from a bigger resolution you end up approaching (or surpassing) 1920*1080*3 samples giving you a much crisper image. Many cameras also have antialiasing filters which may soften the image further when not oversampling the resulting frame. Some other cameras include sensor shifting technology when taking a picture to achieve full per-pixel RGB info (and ofc, more samples = less noise), or just increased final resolution.
Just watched this video from 2018, How well built do you think the Fuji GFX100S? Do they isolate noise from circuitry better then other medium format cameras?
Looks like one more rabbit hole have been opened up on my conventional understanding of cameras; and this is just when I thought that I'm coming to grips with it. I'll definitely refer to this video again as my understanding changes in the future.
Wow this is really interesting! Certainly gets you thinking a little more, I probably should have watched this before posting my last video. Thanks for the info man ☺️
While noise performance may not be impacted by pixel size, smaller pixel sizes means the sensor will very likely be smaller (duh) and thus results in less light gathering forcing higher ISOs, .
You don't need higher ISOs for smaller sensors, the total light across the sensor will be less but the amount per unit area of your scene remains the same
@@DaveMcKeegan While the amount of luminosity per pixel is constant regardless of sensor size, the mere fact of having less surface area for light collection means that smaller sensors will still gather less light and require a higher iso or slower shutter speed to match the amount of light gathered by a larger sensor. Otherwise DSLRs and large image sensors are completely obsolete and unnecessarily more expensive.
@@RonanTetsu Yes a larger sensor will capture more total light, but it's seeing a wider field of view so the amount of light for any given part of the scene remains the same - thus you do not need a higher ISO or faster shutter speed to get the same exposure ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-upjzBSquOYI.html Where the difference in pixel size comes in is actually noise, because a larger pixel will produce a cleaner signal to noise ratio, however higher resolution can down sample some of that noise if not viewed at 100%, hence if you were to compare an A7SIII Vs an A7RIV, at 100% the S would have significantly less noise, however when viewed at 25% they would appear much more similar in noise performance because the extra resolution of the R has filtered more noise out
@@DaveMcKeegan Thank you very for your informative replies and links. Your video on "lens compression" certainly tells me you're reliable; looks like there's much for me to learn (or unlearn) about cameras.
@@DaveMcKeegan So, most of my conversation was meant to target phone image sensors, not DSLRs specifically (though, I guess they're not significantly different?). Do you think or know that pixel size matters in the realm of smartphone sensors, or does it not mean a thing? It's been almost always perpetuated by many that larger sensors and pixels (in the realm of smartphones)=better. But if I had to guess with my limited knowledge, it's not the size of the pixels or sensor, but the actual opening for gathering right?
how much different noise and sensor size in reasonable light conditions is something never brought up so much emphasis is on low light performance which can be balanced with lenses and quality speedboosters also come into it the new ef to eos m by early reviews. On dxo they can be slow what a backlog lol
What about upscale printing? It is not anything out of the ordinary, I've already done it. For example , a 60*40cm print at 300dpi from a camera with 1.6 crop factor with a 20.2 Mpxs sensor, I had to upscale the pixels. In the other hand, If I had a camera with a sensor with the same size but with, for example, 24.5Mpxs, I hadn't to upscale the image. Would I have less noise?
In theory you'd have about the same - the higher resolution would have smaller pixels so would create more noise, but that would then be cancelled out by the down-sample / not needing to upscale as much
This (keeping things cool) makes small body's a disadvantage. But why is the noise performance of the 80D so much worse than the A6500 and why isn't the D3400 better with the same sensor?
I’m a follower since you started and i have learnt a lot from you, thanks for your efforts I disagree with you regarding the pixels size If we are all reducing the image size and it will kick noise then we should all have perfect photos which does not happen I think it is the ability to process image by processor Can you please explain to us the bit depth/rate? Maybe it is the problem?
Downsampling doesn't completely remove noise but it greatly supresses it. Take any high ISO image viewed at 100%, take note of how much noise to detail you can see, then zoom out to fit window, the amount of visible noise greatly reduces. With the argument of pixel size to noise, larger pixels produce less noise, however more pixels allows for better downsampling, the 2 essentially cancel each other out. If you downsample all the images to 10MP then you have the same resolution image off the same size sensor and thus the same effective pixel size
@@DaveMcKeegan Hi Dave, new follower here ;). Would it be perhaps pixel density (eg pixels per/cm2) being directly proportional to noise? Very insightful as someone 'new' (ie less than 3 yrs) to digital photography the back & forth one hears about all this is a wee bit confusing. APSC Nikon shooter btw and quite pleased with results .. but what do I know. ;) Cheers Mate
@@DaveMcKeegan It all seems quite nonlinear to my unknowing eyes. According to your data (which I don't quibble with at all) there's no *direct* proportionality to anything. If there were, at 24P a DX sensor should be 1.5x more noisey, than a 24P FX sensor. No? I agree the data shows a 'correlation' to sensor size. How direct a correlation is not clear to me given all of the underlying tech (processing et al). Won't bother you further mate. Trying to use this as learning opportunity. I'm just going with the bigger sensors are better, done deal. ;) Cheers
Let's say you have an 18 Mpx APS-C camera. Let's say you also have a 1080p monitor, which corresponds to about 2 Mpx. If we take into account that the photos are 3:2 ratio (you get vertical black bars on screen), this means that over 16 Mpx worth of data is simply thrown away each time you view your photos fullscreen. Let that sink in!
although its not really "thrown away" it's downsampled and condensed into the 2MP, so you get not only lower noise but also sharper images than a native 2MP sensor
Not quite true your title, big sensor and low res means also less circuits, less pixels to process, and less heat, the proof in 7iii and 7riii. I think also that manufacturers apply some noise reduction even to the raw so you can't tell why this 2 cameras are better than 7sii, it's because the latter had a worse design or the first reduce the noise better?
The A7III & A7RIII both have BSI sensors where as the A7SII doesn't - the previous A7's and the original R didn't have them either and their ISO scores are all in the 2000's. - It's true that less circuitry will reduce noise interference and may make a marginal difference to noise performance, however the point is to show that in real world usage, larger pixels doesn't have anywhere near the effect on noise that a lot of people believe
Oh you're right about the sensor technology. I don't know how the dxo numbers translate into the real world but between 7ii and 7iii the difference is huge, 1300, about the same between 7ii and 7sii so the only cause must be the pixel size in this case. 7s series was the first camera that was filming cleanly in the dark, so i think it really had the effect on noise the people believed. 7siii will probably kill everybody, even the MF.
problem with that assumption is that most cameras are now ISO invariant and ISO is determined after the A to D converter with the exception of canon. On most modern day sensors the gained remains the same no matter what ISO selected
Think a lot of people do (even I used to). Kind of overlooked that the if you are downsampling your images to say a 8MP file size, then original resolution becomes irrelevant, whatever camera you use is essentially an 8MP camera so the effective pixel size would be the same
Hi Dave, Interesting observation but your analysis is flawed, based on your anecdotal camera comparisons. First you need to understand how a camera sensor works in order to properly answer this question (refer to Shane McLachlan's explanation below in the comments, he has more knowledge on the subject than I). Also, you are vastly underestimating the weight of variables that effect camera sensors performance other than pixel density, such as sensor technology, processing power, software design and its efficiency. Other omissions in your video analysis that you did not consider are; 1) the price of each camera, which typically has a positive correlation with sensor performance, as a result of the implementation of a more expensive (better performing) sensor and faster computing processors; 2) the date each camera was released, since sensor technology, software and processing power are continually improving; 3) and the intended purpose of each camera, which inherently effects how a camera manufacturer designs and optimizes it (e.g. video camera, general purpose/stills camera, hybrid camera, sports/wildlife camera, landscape camera, portraiture camera). Also, when comparing low light ISO scores from DXOMARK between different cameras and sensor sizes, why did you not include the best performing MFT camera? You included the best performing FF cameras... The Olympus OMD EM1 mark ii has a sports (low light ISO) score of 1312 which is far better than the GH5's 807. I guess it would muddy the potency of your comparison of sensor size performance, since the EM1 mark ii scores better than two of the larger sensor APSC camera examples you provided. Panasonics G9 will likely score similar to the EM1 mark ii once DXOMARK gets around to testing camera sensors again. As for down sampling images, you are correct, it does in fact lower the amount of noise in the image. However, not to the extent that it would perform on par with a larger pixel camera (lower resolution, same sensor size), given that all variables mentioned in my first paragraph were equal (which is very improbable). FYI, DXOMARK already controls for noise reduction in their low light ISO scores by down sampling each raw photo to 8 MP. This standardizes their low light ISO performance scoring method, allowing for comparisons to be made between different sensor sizes and pixel densities.
Honestly Jay I just completely forgot about the Olympus, it would have actually been ideal to include because it would further demonstrate the point about pixel size.
@@DaveMcKeegan What it shows is there is a lot more to camera performance than sensor size and pixel density, hence all the other variables you should be considering when comparing sensor performance. However, the only reason sensor size has any bearing on image quality is a direct result of pixel density, its just hard to see that in your comparison because you do not take into account many of the variables mentioned in my comment above. Camera sensors do not break the laws of physics!
I'm not suggesting they do, I didn't say that smaller pixels give the same noise performance - I even said in the video that when viewing at original size then pixel size has a big effect, however very few finished stills are viewed at 100%, the moment you downsample you essentially cancel out that additional performance gain. If you scale 2 different full frames and scale the images down to 5Mp, you are basically comparing 2x 5MP sensors
@@DaveMcKeegan You just created a paradox. Your theory would allow for camera manufacturers to infinitely increase pixel density on their sensors, at no cost to performance for an infinite amount of noise reduction when scaling down an image. When you pixel density is higher, each photosite takes in less information. Yes, there are more pixels but in low exposure situations they would all have less information to begin with, so averaging pixels together would not improve the quality of the photo nearly as much as you seem to claim. A sensor with larger photosites (lower density) would capture more information in the same low light exposure allowing for more information to be captured by the sensor having an overall better quality photo.
Dude. I think your math is wrong. 70.5 micron pixels in the a7sii?? That would be HUGE! You're forgetting to square root the quotient of sensor surface area divided by pixel count. 847mm^2/12mp=0.000705mm^2 or 70.5microns^2 √70.5 = 8.4 microns across.
If pixel size/density didn't matter, the Arri Alexa would be cramming >100MP in their cameras right? If pixel density didn't matter, the A7S2 wouldn't have ripped the A73 to shreds in a low light scenario. There has to be a tradeoff. As someone said, there ain't such a thing as a free lunch. If it didn't matter, manufacturers would be stuffing their sensors with 100, 200 megapixels. Because larger numbers always entice the general public who don't know what it means.
There is a trade off in noise when viewed at 1:1, but realistically for stills most people never view a finished image at 1:1 - anything less than 1:1 and you begin downsampling which filters out the noise, which is why at the same size photos the A7SII & A7III & A7RIII have very similar levels of noise. The real trade offs are things like speed, adding more megapixels means more information that needs to be pulled off the sensor which slows things down ... which is a bigger problem for video shooting because it'll create more rolling shutter
Oh my lord, even as a photographer and a technophile I have never once heard a person even bring up that idea. From my perspective and my self-research, ISO noise/noise has always occured because of the Computer Processing Unit and Random Access Memory, as they were not powerful enough, had miscalculations or not a well-enough coding algorithm implemented into the cameras.
BTW sorry for the off topic but I'm analyzing changing my set of lenses but afraid that I'll have gaps on my focals. I want to know what's better between using an extra sharp lens (Sigma 50-100) and then crop or using just the (Canon 55-250) is there any way to do a mathematical comparison between both lenses at, I don't know.. 200mm? Thanks
There isn't really a hard and fast rule, it depends on the individual lenses. Better quality lenses will allow you to crop in more before showing drops in image quality.
The mpx should give you a rough guide, however sharpness will vary throughout the zoom range so depends if dxo score the lenses at sharpness focal length or maximum
But smaller pixels means smaller noisey pixels and more pixels means you can average the noise out. Yes larger pixels produce less noise but that is only visible when viewing images at 100% If you view high iso stills between an A7S and an A7R, when neither are at 100% the level of noise appears very similar
A7RIII - 3523 Vs A7RII - 3434 A7III - 3730 Vs A7II - 2449 Vs A7 - 2248 (all same resolution but the A7III was the first of those to use a BSI sensor which makes a big difference)
When viewing at 100% or with standard video then you will see a difference because you are viewing at a pixel level Once you start down sampling the noise starts to get cancelled out - more pixels will mean smaller pixels so more noise, but more pixels is more down sampling so it cancels itself out
Always impressed by how much thought you have put into the physics and technical aspects of photography rather than just blind acceptance of accepted dogma. Very refreshing.
Yes arguably taking a picture with a cold sensor and taking a picture after prolonged video recording would result in a difference in noise performance.
So, it means low megapixel cameras(GH5s, A7s...) are for video works because low megapixel could make frame rates higher or color depth deeper with same processing power, not for the low light performance. Interesting and informative!
Seriously? How is noise measured? Is this claim just based on opinion only? Or is this just based upon mathematical constructs, or what? I make no claims of being such a brainiac. I for one happen to believe that a larger sensor surely makes digital noise appear less noticeable, and I own both a DX format camera, and a full frame camera. I'd love to also see a comparison done using a cheap phone camera with a tiny imaging sensor. It also matters if you crank the ISO up into the stratosphere. I examine my photos on a reasonably good quality monitor. But if I made even huge enlargements, actual prints, then I doubt I'd see any difference.
Noise in an image is random, however DXO measure the amount of noise in the signal which is a lot more accurate. Larger sensors do have less noise but this is about more megapixels, if you aren't viewing images near 100% then 2 sensors of the same size with different megapixels won't show different amounts of noise
Lots of people shoot astro with Sony, star eating is actually a thing of the past with this new generation of cameras. Find the comparison video someone did comparing the A7III Vs 5DIV for astro
@@nordic5490 If so; can you explain why the film "An inconvenient truth" was banned by court in UK due to having "nine scientific errors"? news.bbc.co.uk/2/hi/7037671.stm Quick quote from Wikipedia, but there are other sources: Effects of sensor size The size of the image sensor, or effective light collection area per pixel sensor, is the largest determinant of signal levels that determine signal-to-noise ratio and hence apparent noise levels, assuming the aperture area is proportional to sensor area, or that the f-number or focal-plane illuminance is held constant. That is, for a constant f-number, the sensitivity of an imager scales roughly with the sensor area, so larger sensors typically create lower noise images than smaller sensors. In the case of images bright enough to be in the shot noise limited regime, when the image is scaled to the same size on screen, or printed at the same size, the pixel count makes little difference to perceptible noise levels - the noise depends primarily on sensor area, not how this area is divided into pixels. For images at lower signal levels (higher ISO settings), where read noise (noise floor) is significant, more pixels within a given sensor area will make the image noisier if the per pixel read noise is the same. For instance, the noise level produced by a Four Thirds sensor at ISO 800 is roughly equivalent to that produced by a full frame sensor (with roughly four times the area) at ISO 3200, and that produced by a 1/2.5" compact camera sensor (with roughly 1/16 the area) at ISO 100. This ability to produce acceptable images at higher sensitivities is a major factor driving the adoption of DSLR cameras, which tend to use larger sensors than compacts. An example shows a DSLR sensor at ISO 400 creating less noise than a point-and-shoot sensor at ISO 100
Great attempt at debunking one of the most common ignorance-induced myths in photography. I would like to make the point that I missed in your video (maybe it was there in some other form). Usually this is not really talked about because most people are not scientifically trained. Noise is an image-level metrics, not a pixel level metrics. Each pixel (or, more technically accurately, sensel) gets hit by photons, light energy gets converted into electrical energy and the sensel reports the total value of this energy for the duration of the exposure to the processor. If you have a single value, how can it have a noise? Can number 1 have noise? Or number 879086246? Of course not. Noise arises when the numbers reported at each pixel deviate a lot. That is why when the signal is high, those deviations remain very small - like blue sky on a sunny day. Also, when values are supposed to be very close, it is easier to notice the discrepancy - that is why noise is most obvious in areas where fine detail is absent. So to speak for noise of individual pixel is simply nonsensical. If people mean noise when image is displayed at 100% resolution, that is something different. That metrics is not nonsensical, it is just silly.
Yeah I got there in the end 😂 Firmware certainly plays its part in reducing noise which is why I wanted to include the 3x A7's, similar firmware but huge differences in resolution should have made a noticeable difference but it didn't
