You're missing something. Some sensors do actually change at a certain iso. So my Sony A7iii switches to a different mode when you hit 640. It amplifies the sensors signal via analog gain which produces less noise. So the camera is isoless...as long as you're in the same "range". 500 vs 100 would be the same. 3200 vs 640 would be the same. But 3200 vs 100 would not be the same.
www.dslrbodies.com/cameras/current-nikon-dslr-reviews/nikon-d850-camera-review.html This article mentions a dual iso. I'm just saying it's worth mentioning that it can be a variable. Then people can decide what they want to do with that info! Haha
All cameras are different but from the linked graph it appears the D850 has two levels of ISO invariance, one between 100-320 and one from 400 upwards (photonstophotos.net/Charts/RN_ADU.htm - and chose D850 on the right). Some cameras exhibit ISO invariant behaviour from 100 upwards, some like tha A7iii and apparently D850 have two levels of ISO invariant behaviour. Some like the 6D only exhibit ISO invariant behaviour after ISO3200. Probably the best write up about ISO invariance can be found on lonelyspeck or photographylife. www.lonelyspeck.com/how-to-find-the-best-iso-for-astrophotography-dynamic-range-and-noise/
@@jeffluo9591 I'd like to know more about that, and I seem to be reading conflicting numbers here in the comments, while a quick search on the internet only yields articles with vague claims. You claim the thresshold is at 640, Alyn Wallace claims the threshold is at 400. Have any sources so I may know where the threshold actually is? (interested on the D750 and D850 values)
Tony gets criticized a lot for some of his opinions, but I’ve found that he’s usually able to back up his claims in pretty reliable ways. As a digital media teacher whose classroom is right next door to a traditional film classroom, I can confirm that iso from film to digital is a mess of confusion.
Read my comment above, unless you have an "ISO-invariat" sensor Tony is very wrong. If you expect to see what he states, to say with the just announced Canon RP you would be very disappointed.
@@armandot9137 even with an ISO invariant sensor there is more bit depth in the highlights of a raw file than the shadows, if you're pulling information out of the shadows you'll lose some amount of detail compared to if you stored them in the highlights of the file by using a higher ISO. on the other hand using a higher ISO costs you dynamic range.
@@armandot9137 Regardless of the method used to increase and decrease the sensor sensitivity, ISO standards should yield the same results in exposure (within a margin of course).
@@tobiasyoder it is indeed but we're far more scientific in our approach (and have a million more reasons to understand ISO properly). But we're kinda fed up of people saying false statements like "higher ISO = more noise". Nice to see people are waking up to ISO invariance though, it's highly useful knowledge
Lol hoping astrophotography sounds as cool as astrophysics. No different from food photography. Anyway, from what I've gathered, ISO is just gain applied after the exposure.. however the gain applied by the camera's amplifiers in the signal chain is different from the gain you can get from software. Maybe that's why the exposures in this video looked different.
@@Triple070007 yep it's gain but the difference is that some cameras induce noise into the image after the amplification gain in camera. This is why there's a difference with boosting in post with some cameras. Cameras that don't add much noise after the amplification in camera are those that behave iso invariant so you can boost in post and get a similar result
Some modes of applying exposure gain are better than others. People super into photography tech are still adorably unaware of how anything about cameras work. News at 11.
@@AlynWallace same boat here, seen too many time people doing astro saying "I kept ISO low to keep the noise low" while they actually destroyed the image. Ironically enough an article was recently published on Fstoppers explaining ISO the wrong way :)
Some of the noise comes from the rounding of the number. In a 14 bit raw file, the intensity of a pixel changes from 0 to 16384, so a correctly exposed picture's rounding error is almost negligible. However, when you underexpose by 5 stops, you effectively have 8-9 bit raw file bit intensity varying from 0 to 200-500, where shadows of this image have a small number of intensity. Then the rounding error becomes significant.
Few years ago I’ve read a research paper about photo sensors. The skinny of it is: sensors do have constant dynamic range but photo cameras also have hardware noise reduction that cleans up RAW data before saving it. Some dynamic range is cut on the bright side in the process, that’s why you can’t recover overexposed high ISOs. Extended ISO range is when hardware NR is disabled or inactive so you basically get same results as Camera RAW exposure slider. And there is Red cinema cameras that function exactly as Tony described - if sensor itself wasn’t clipped and there is still signal in shadows then you can totally recover anything. However Red has new Gemini camera with dual ISO and it’s low light ISO setting does have hardware NR, so you get lesser dynamic range, but also much cleaner image with higher overall exposure
At a basic level of understanding, the ISO is just brightening the image; But for a better result, its better to do this before committing the data to file than after. A digital file format has limitations at extreme dark or light dynamic ranges, because the information is quantized in to a small part of the dynamic range. The result is that the file cant hold as much information from a dark image than a correctly metered image. This I think is the main reason for the differences.
You are most right about what this video couldn't figure out. Nature has almost unlimited colors but digitization must limit the color palette one way or another. It's actually a form of compression artifacting but in a small scale since those files are technically(by name) uncompressed.
It's a lot like recording audio; Good Level + no post gain = better quality and less distortion. Low Level + post gain = worse quality, background noise, artifacts.
I agree. I do sound recording/mixing. Getting a good level into your mics trumps all just like getting good light levels in photography. It means you can drive your preamps lower (equivalent to using low ISO). Analog gain circuitry like mic preamps adds noise especially if pushed too far. In that case i opt to stay a bit lower and use post digital gain later if necessary. It seems to keep things cleaner.
5 років тому+1
This will practically change how I go about my shoots. I used to be scared of underexposing, now I guess there's one less thing to worry about in the field. Of course we as photographers should do our own further testing with our own camera models. I'm gonna do that right away after this
I think you guys made the same point but maybe I'm confused a little. In your first test, if the D750 and D850 shows the photo is overexposed @ ISO 400, doesn't it mean that you can get the same exposure as the D1H @ a lower ISO?? Hence the selling point!
Tony was arguing that you would take a photo with an old camera at iso 3200 and a new camera at iso 3200 and notice there is less noise in the new cameras image at the “same” iso. While in reality it might actually be shooting a darker image at iso 1600 but it changed the number to 3200
Something is missing here, the D1h is using CCD sensor while others are CMOS sensor. So I reckon that the ISO calculation maybe different because of this.
Who says the Lightroom algoritm is identical to the ones in-camera? People tend to take LR as THE ONLY and CORRECT post processing tool, no matter how many times we have seen that LR does not handle a specific raw file from a specific camera any good ? Heck they seem to still use the dreaded recovery-tool before you even see your photo...
What Tony says if I understand it correctly is that if that the image is pushed 5 stops in the camera the images would look exactly the same?? All software process the images differently. Lightroom would not do it in the exact same way as the software in the camera.
Magnus Eriksson I want to try this but Nikon’s in camera NEF editor only pushes an image 2 stops. Maybe their proprietary NEF software ( is it still capture nx?) would do the same thing? -P
Tony just exposes his ignorance when he claims that all cameras exhibit "ISO invariance". ADC (analogue to digital conversion) is handled differently by different sensors. Some use analogue gain in which case Tony's claims fall flat. Then there are the modern dual gain sensors where ISO invariance holds but only in two separate ISO ranges. This stuff is really not complicated.
There is one important comment to Tony's video to pay attention to, that the ISO is an analog gain applied to the signal (which is done as the gain on a stereo, using voltages) before it is converted to digital. Once the signal is digital, you have lost the fine details, and boosting it in post will cause "quantization errors", which looks like noise. Example: An analog signal can represent any value between, say, 0 and 1. It can be 0.1, 0.123 or whatever. A digital signal can only be either 0 or 1. So when the analog signal is digitized the values are rounded.
First of all.. ISO in sensor is obtained by changing the Vcc voltage of the sensor. Although it's not canged for every ISO setting rather is decided in bands, for eg. 100-400 have the same input voltage of the sensor and the rest is processed by the processor. Second, the Exposure settings in LR or PS are complex algorithms simulating exposure equivalence but is not eqivalent. Third, in the internal processing of an image some other things are applied, like dark current clipping, witch in Nikon cameras is pretty hard, and this is the reason untill d850 and partially d800a the Nikon cameras where not the one used for astrophotography. Dark current clipping is omitting the darkes informations in the image, to be breaf. Fourth, the RAW file is nothing like the the true image comming from the sensor, it is heavily processed, not only in sharpness or contrast... The real image from the sensor is mostly magenta in tone and you would never be able to match it to the scene by hand in PS.
No. The sensor's sensitivity has *nothing* to do with its digital supply. Analog gain is a thing, but you have no idea what you're talking about. Multiple gain architecture is only used on some sensors, others only have one analog gain set in silicon. The real image from a sensor, when debayered, will look very green, not magenta.
@@Spirit532 I don't understand what you say about the 'real' image which would look green. The bayer matrix of the sensors has twice as much green points than red and blue, so what would be the real image? What I understood that the human eye is extremely good at distinguishing between slightest differences of greens but not as good on other colours, thus sensors use more green elements than other colours. And of course since three colours channels are recorded and you want to keep the rectangular and repeating matrix one colour has to have more sensor elements than the two others.
The difference you found is actually the process of of raw processing the "High ISO set" image to shift the "Chromatic Noise" to "Luminance Noise" which is more pleasing to a photographer because it looks more acceptable and filmic the "silver hallide clumping" yumminess of film grain is not always bad. It's also a way of mitigating huge file size as the resulting colour noise would make the file incredibly huge the colour averaging of the noise make it more manageable to save.
Great videos by both of you guys! So, we don't and will probably never really know if the "just boosting gain" theory is exactly correct. Even if the gain theory is correct as to which Tony suggests, the way Adobe Camera RAW interprets boosting 5 stops to get to 3200 ISO compared to say a Nikon RAW file "Before" it becomes a NEF file COULD be different. Hince, why you are seeing more grain in the 100 ISO boost shot. Either way, this is super interesting stuff and I am glad there are people like you guys and Tony who make videos about these subjects!!! :D Also, using your REALLY old Nikon might be too far apart generation wise. The sensitivity of the first generation sensors could have just been really bad on those really off compared to what we have now. But that is hard to say. Would be cool to test with a D700/D3 VS a D610/D750 to D810 to D850.
Jim Boomer this is what I was saying off camera. I wanted to test it by boosting the 100 iso file in camera with Nikon’s NEF processor. Unfortunately it only boosts the EV by 2 stops. -P
Exactly - I suspect the files are different only because the algorithms used to boost the exposure in the camera are different than those used by ACR. So not necessarily more noise, just different noise. I think we DO know that it's just boosting gain - we just don't know exactly how each camera does it. The sensor would have to physically change for the actual sensitivity of the photosites to change. I also think older cameras are definitely "less ISO-less" than newer cameras.
These are not "great videos" at all. Both authors don't seem to know the first thing about ISO on digital cameras. Look up "analogue gain" vs "digital gain". Look up "dual gain sensor". Look up "ISO invariance" and understand when it does *not* apply.
This video was really quite well done. I really liked how you were so humble and not trying to find fault with Tony or anyone else. You're a class act!
Looks like you proved BOTH of Tony’s points and perhaps were just confused on the first test. The newer cameras make the same exposure look “brighter” (hence fake better ISO performance). The second test is also proven, and the “noise” difference Lee points out is splitting hairs and easily attributable to differences in the gain calculations between Nikon in-camera vs Photoshop in-computer.
I think you have it backwards. If ISO 6400 was actually ISO 4000 in order to make ISO 6400 look better then ISO 400 would be actually ISO320 if the pattern stays consistent. What I have found with all my Nikons and m4/3 cameras is that ISO100-800 seems to be the same across most cameras and only when getting to ISO 1600 and up do the shenanigans start where it reads a higher ISO than what it actually is. If ISO 12,800 is really ISO 8000 then a shot at that setting would look pretty good on a FF camera when the previous model maybe had the same look at shows ISO 10,000 but was actually ISO 6000. My m4/3 looks great up to 3200 and still very good at ISO 6400. If I had a light meter I could see what the meter shows I need at that ISO to see if the camera is lying or not. When I tested high iso, I just took images at ISO 1600 and up in 1/2 ISO steps and was able to see where the quality was too low for my standards for whatever size print. I tried with my Nikon stuff and found the variation among the 3 cameras I had but I didn't test the same exact settings on the same studio lit item with m4/3 to directly compare the same lighting and see how it matched up ISO wise.I have a D5300 for the Nikon 10-20mm, I will have to try it against my G7 and G85 and see if they are all the same at the same high ISO.
Yeah and noise reduction filters may have been set differently in the RAW file after the picture was shot. Nikon Picture Control can even affect some of that noise. For example, higher contrast setting..? I saw Tony's video before I saw this one a day or so ago. I agree with you, it would seem having more brightness at lower ISOs would be an advantage. And so iso 100 is actually more like iso 50 like what Tony says... I think that he proved his claim.
I'm confused. The shot that was pushed five stops clearly shows much more chroma and luma noise plus the funky white dots.How does that prove anything except shooting at 3200 produces better pictures?
For the iso invariant test, your software matters. Capture One will give you a better result. Also, you get more highlight details in return, when you do that kind of boost, that's how you can get the "Film look".
@@ARMAJOV Indeed, but Capture one does some things automatically if you don't disable it. You can get (nearly) the same in Photoshop, it is just not the same set of defaults. In the tests in this video I wonder if Nikon does some high-ISO noise reduction like Sony does. If this is not deactivated, of cause there is more noise in the low-ISO picture. Tony did not mention those details. (According to some camera tutorials I think he must be aware of this in general. Maybe he left it out intentionally.) But the claims are more or less correct. The picture is the same within the sensor, the rest is signal processing. (Just ignoring possible dark frames or combinations of more than one picture in camera.) Reducing this topic to the exposure setting only is a bit too primitive thought. Lee is really no technician. @Lee, if there is a setting like Sonys high-ISO noise reduction, please disable it and re-shoot your example.
@@ulrichsiebald144 that was kinda my thiughts also. However the iso 100 pic has a ton of noise. Just different noise. If noise reduction was on its pretty crappy.
Watching both yours and Tony's videos today was fascinating. You didn't so much "bust" Tony's claims, as essentially prove the same things. Sure, there are very subtle differences, but overall I think you guys both nailed the basic concepts. But, despite all the techy stuff, the thing that impressed me the most about your video was the clear respect and admiration that you gave Tony, even though your conclusions were slightly different. This is EXACTLY the kind of robust debate that makes it a win-win for all of us. Thank you so much for being firm, clear, and very gracious.
We need to use a scientific standard to sort this out and call out fowl play. The lumens could be measured from glass to sensor and sensor readout directly but we would need an efficient method of transfer after that to judge final output, manufactures probably wouldn't like this plus the different software engines they are running will have a part to play as well. Not too mention, the frequency range of all light gathered and the levels from within would also need standardisation. A bit like measuring headphone performance (actual studio quality, not prosumer, rich kid or gamer BS).
If you look on DXOMark, they actually test sensitivity vs advertised ISO, and from what I recall, there's quite a bit of variance, but most cameras are somewhere around half a stop less sensitive than their nominal ISO values suggest.
If you understand how photo data is saved as a matrix, you understand why there is more "pepper" noise when you bring up the exposure for the ISO 100 image. Depending on how many bits of data are saved to each pixel can vastly change the results. High ISO settings automatically multiply the pure raw data before being saved as a 10-bit ir 12-bit. Amplifying the ISO 100 image multiplies the already "lossy" 10 or 12-bit value AFTER being saved as a 10 or 12-bit.
Hi Fstoppers .. I don't think you found evidence for "the opposite to be true" I think what you discovered in your old vs new Nikon teest is actually the fundamental problem Tony Northrop refers to. The sensor technology between the old and the newer cameras improved so much that the older camera needs to boost the signal (gain) significanly more than the newer ones. Increased gain generally leads to more noise. So the improved sensor technology allows you to shoot the same picture at a lower gain and therefore noise. Regarding ISO number inflation, there might be some truth to Tony's observation, until there is a clear standard/reference it can and will be used to make the product look better. Bigger Numbers make it look better for sales... same as GHz on devices, HP on cars etc. ... Regarding your second test Yes there can be some "processing" going one before the image is stored ... you need to digitise the analog signal Every pixel in the end has a limited capacity of storing different information. Lets assume there are only 1024 brightness levels Correctly exposed you will used camera gain and all 1024 Levels, but when underexposing 5 stops you cram all the information in just 32 levels Because the gain gets applied after the values are digitised it is possible that this causes artefacts Analog | 5 Stop Gain || Digital Value | 5 Stop Gain 7.968 | 255 || 7 | 224 8 | 256 || 8 | 256 8.0312 | 257 || 8 | 256 While the Analog values are actually very close to each other the post digitised values can be the same or very far apart In an ideal world where the RAW file is able to store infinitely/sufficiently precise, there is no difference between the two For now, using ETTR you can use the information capacity within a RAW file as much as possible. Also high information RAW formats today allow you to be off by 2-3 stops, without loosing too much informatinon .. and as always ... your mileage may vary ...
For any given camera, ISO is metadata-only (so doesn't matter if you change it in camera or in post - also called "ISO invariant") for a given range or set of ranges - usually above a certain ISO. For example the 5D MarkIV is ISO invariant between ISO400 and anything above - but below that it is NOT invariant. the 5D Mark II was not ISO invariant until, if I recall correctly, well above ISO 1600. Red cameras are purely ISO invariant, as is the Arri Raw codec. You have to look up the model of camera you are using to find out the range or ranges of ISO invariance to know. Testing one camera only tells you about that one model, and even then you would have to test all ISOs to find the range(s) of invariance.
You should retest using the same tstop and lens across all cameras rather than going off the fstop. Usually fstops are off on most manufacturers, check dxo mark.
In video this is a bit more known because videocamera’s often use gain measured in decibels. Alister Chapman had some interesting content about that. In the end, amplifying the signal on the sensor level will probably always be better then in post but the difference is small. RED camera’s even use ISO/gain as a metadata setting, you can set the iso in post!
When it involves skin tone, though, I think that ISO-100-pushed-5-stop photo will show unnatural look. I would shoot at ISO 3200 when shooting people in this setting.
It would have been a more interesting test if it was only 2-3 stops underexposed. I bet they would have recovered almost perfectly in that case. The fact you can recover five stops and even have a somewhat usable image is simply mind blowing.
@@elvirredzepovic6898 funny thing I make one people test from his d810 which is supposed to be iso invariant at 400 iso. And he said yes it's more magenta if you boost 640 iso to 6400 iso. Of course this moron didn't set the same white balance. So sometime people Cannot test correctly
That thing about those white dots is really important. I use an entry level DSLR and I have encountered huge problems with underexposing my images by mistake. Once when I underexposed my images a lot and brought them back in Lightroom some of the areas which were black turned completely white while editing. Lightroom just went crazy. If I underexpose by one stop it's fine but when I try to push it further the editing part gets tricky at times. ISO might not seem like a big deal after all, you just need to get it around the correct value but when it comes to us - entry level camera users or crop-sensor users in general the difference between setting the correct ISO might be huge.
Tony failed to mention the sensor needs to be iso invariant you need to check the list of iso invarient cameras. Most Sony’s are. Canon isn’t fuji xtrans is. Also there are two groups. Iso base -400 then iso 640 to 6400. My Sony a7iii is invariant as are my fuji. But the 5dmk4 isn’t. I didn’t think the Nikon was either. I’ll agree that applying gain in post as opposed to in camera has its benefits for example retaining highlights and better colours. But the in camera gain seems better than Lightrooms. I did it and explained in my video a few month back. As far back as the fujifilm xt1 was invariant. Weird that canon isn’t. It’s ideal for timelaps shoots as you can underexpose and retain better highlights and lift the exposure instead of clipping when it got too bright for example 😊👍🏻 there are a few cameras on here that are and are not. Check it out 😊❤️📷 improvephotography.com/34818/iso-invariance/
the 5Dmk4 is "partly ISO-invariant". "The 5D Mark IV isn't entirely ISO-invariant: pushing an ISO 200 underexposed by 5 stops by 5 EV in post-processing yields slightly higher noise levels than a native ISO 6400 exposure. An ISO 100 exposure pushed 6 stops fares even worse. However, above ISO 400, the camera does, for the most part, exhibit ISO invariance, meaning that you could underexpose a traditional ISO 6400 exposure by 4 EV by shooting it at ISO 400 (while maintaining the shutter speed and aperture for ISO 6400), and then raise exposure 4 EV in post. This technique would afford you 4 EV of highlight headroom, with little to no noise cost, relative to shooting at ISO 6400." www.dpreview.com/reviews/canon-eos-5d-mark-iv/11
Even newer cameras apply analog gain to the sensor signal before running it through the ADC when you turn the camera ISO up. To a point, analog gain before ADC will result in a cleaner image as a significant amount of the noise you see comes from the ADC and a larger signal into the ADC is a cleaner conversion. The larger signal can be either more light, or more analog gain. More light is cleaner than more analog gain, but more analog gain is cleaner (by a lot) than gaining the signal up after the ADC (digital gain).
In your test with the three cameras, isn't it the oldest camera that has the darkest picture at the same ISO as the newer cameras? This result then isn't "the opposite". You have to lower the ISO setting of the newer cameras making it appear that they get the same exposure at a lower ISO.
This is interesting. I'm an old film shooter, did a lot of very tight technical exposures. What I noticed when I started shooting with a Nikon D2X is that it matched my Minolta IV meter (incident) exposures and was itself accurate enough to use as a meter. I'd set the D2X on P, then transfer the settings to a 4x5 or 8x10 camera (adjusting for bellows of course) and still get a very accurate exposure. That was 10-14 years ago. I also started using my D2X for B&W film settings but there's a lot more latitude shooting negative. When I first started shooting jobs that were part 4x5 /8x10 film and digital DSLR, the settings I got from my light meter worked well in my DSLR. So I would say that through the Nikon D2X ISO matched the long standing criteria. It doesn't surprise me that some companies might fudge this. I don't think Nikon or Canon would do this on the pro-level DSLR gear. Your 3200 then 5 stops underexposed really surprises me. What is the range of a camera sensor? A typical film shot is if I remember has on average about 7 stops in range in the image. This doesn't mean you can shoot + or - 3.5 stops, this means that at the bottom end is almost black (DMax) and at the top end is almost pure white. A difference in 3/4 of a stop is noticeable, less is pretty much the same (in transparencies) three shots 1.5 stops different, even in a color negative film means two of them will be useable. If you expose 5 stops under? In transparency film that is pure black (if I had a reflection that was - 5 stops, I could see the reflection, but on film there would be no reflection). I'm curious as to what the actual range the sensors capture raw data, that's the real determining factor. Good video.
Indeed. More precisely, according to Wikipedia: >> ISO is not an acronym. The organization adopted ISO as its abbreviated name in reference to the Greek word isos (ίσος, meaning "equal") But in the end, hearing it as being pronounced "I.S.O." here and there is really not a reason to go nuts ;-) Go out for shooting, that's better for the health than to discuss endlessly the right pronounciation of the 3rd member of the exposure triangle. Oh wait, it's not even that; Ken W. would bash the sh** out of me if he read this technically false statement...
@@korm87 yes previously asa was used and that was an acronym. Iso is not an acronym though. www.iso.org/about-us.html Straight from the company's own website. They say it themselves the founders used the word iso taken from greek word isos meaning equal.
Jeezus you guys, talk about splitting hairs. The organization is in fact an acronym I.S.O... But in order to keep the rating acronym consistent across different countries, they came up with some silly excuse about the greek meaning of Isos... Again, literally just to make sure every country said their ACRONYM correctly. Actually getting annoyed and correcting that level of minutia frankly makes you look dumber than just using the acronym everyone understands.
I love it when my favourite channels reply to each other and even challenge claims of others (in friendly manner). No matter who is right in the end, it puts out quite a lot of interesting information out and a few forth and back discussions generate more tests and claritications.
You are not even convinced of each other's explanations and arguing who is right or wrong. This type of info could be answered only by manufacturers. So this is all nonsense. Goodbye
I could be wrong about this, but I thought I had heard from an engineer I was talking to at NAB last year that ISO numbers for digital sensors are all relative to the sensor itself. In other words, a sensor will have a “native” ISO (usually around 400) that represents the raw data from the sensor with no manipulation, and all other numbers just indicate the factor that native ISO has been manipulated. For example, if the native ISO is 400, 800 would simply be whatever manipulation would be needed to increase exposure by one stop, 200 would be manipulated to decrease by a stop, etc. In that regard, it would make sense that a newer camera with a better sensor would capture more light at its native ISO, thus causing the image to be overexposed when using the same settings as an older camera. It would also explain the differences in results as every sensor would have its own native exposure sensitivity. In regards to pushing ISO in post as opposed to on camera, I always thought most DSLRs do some low-level noise reduction at higher ISO settings? That would explain the difference in results when pushing an ISO 100 RAW image taken with the same settings as a properly-exposed photo shot at a higher ISO. Anyway, just my $.02.
The lost dynamic range is a significant thing to keep in mind with modern cameras. As you increase your ISO, you start clipping highlights within the RAW file. Once that file is saved, and that pixel is registered as a full well, there is no longer any data to be recovered. So, if you shoot a file at ISO-3200 to get detail into the deep shadows, you will lose, beyond recovery, the details in the highlights. The reverse is not true, as you have just demonstrated. Even 5 stops of push (Which is a damed lot!), does not cause any significant degradation of the file quality. Hence, the old rule of "expose to the right" is defunct.
Thats not exactly true. Certain cameras actually gain highlights as you go up in ISOs. For example. The blackmagic pocket Cinema camera 4k actually has more stops in the highlights as you go up in ISOs, then it resets when you hit the 2nd circuit of the dual native ISO
Exposure to the right is not defunct at all, it's just that changing ISO has no effect on exposure. Exposure is a function of scene luminescence, aperture ratio, and shutter speed.
Same comment I made on Tony’s. Since ISO is gain you can not isolate the result from signal to noise. It can vary greatly, just as in audio equipment with the quality of the amplification (hardware) and the case of the digital the, software.
I have noticed that some people are using the term ISO Invariance to mean "mostly the same" as opposed to meaning "Identical". Mostly the same, as I understand it, means that that the change due to gain ( before the ADC) is similar to the change due to doing some math in software. Which is a good thing. I guess it just means the engineers making the hardware and firmware have done quite well at reducing noise where it can be reduced.
@@bassangler73 its international organization for standardization. Iso is a 3 letter word the founders of the company used, which comes from the greek word isos, meaning equal. This information is all public knowledge and free for anyone who cares to take the time to go to the company's website and look for themselves. Or you could call them and hear them answer the phone amd see if they say i.s.o. or eye-so (spoiler alert, they will say eye-so). 🙂
Boosting in post is often indistinguishable from raising ISO in camera for under exposed images. BUT of course blown highlights can never be recovered in post so dropping the ISO must do something before the raw file is baked.
Uggg,....all image sensors are pre-amped when they are converted to digital. That conversion is set by the camera company to be whatever "ISO" they want their "0db" to equal. (some cameras have TWO "0db's"...low 0db and high 0db) and Remember...all raw sensor data is saved at "0db"....ONLY. There is NO USER ADJUSTABLE GAIN APPLIED TO THAT RAW COLLECTION. This only happens whee the raw data is assembled into a .jpg. The raw data ALWAYS stays at native "0db". If you add +12db to your .jpg, than a "+12db" FLAG is added to the raw metadata. It's only a FLAG given to your RAW reader to apply a default +12db on RAW assembly. The RAW data in the file still exists at "0db"...always. Gain is a destructive process and that's why it's NEVER calculated into RAW's 0db status. This is why raw has the highlight recovery ability that it does....because you CANNOT CLIP IT USING CAMERA GAIN LIKE YOU CAN A JPG!
My theory for why there are differences is that the sensor data gets first multiplied by the gain setting and than is stored in the same way as it gets stored in JPEGs, as square roots due to the way human perception of light works.
The amplification value of the chosen ISO is added before the digital conversion of the signal, and not afterwards, like changing the exposure value in Lightroom would. This decreases the noise in the image.
Very good addon video, wish that someday there would be a frotalk gathering with fstoppers, tony& Chelsea and may be Ken Rockwell too. It would be fun to see you all together
Tony used Lightroom, you used Photoshop. LR has noise (and color noise) reduction sliders which if you have used them, then the image would have been exactly like the one taken from Nikon at ISO 3200. The camera simply applies some noise reduction techniques when shooting at an ISO higher than 100, in rest is just software gain.
Critical Point But does it apply noise reduction before writing the NEF file? I was always told raw files don’t have any noise reduction/sharpening/shadow recovery etc. -P
Critical Point - Then doesn't that prove his point, when treating both files the same way (meaning, not doing color noise reduction) doesn't that reflex that the files are not identical?
@@FStoppers Check out some of the forums on dpreview or photonstophotos, Nikon is known for baking in a level of noise reduction in blacks in the raw file itself.
Really well done! This strikes me as a good example of how the scientific method should work. Tony posited some hypotheses, you guys test and find your results, then the data are what they are. I really appreciate that over stupid, drama-based flame wars.
The iso 100 exposure's much shorter shutter speed will mean a lot fewer photons hitting the light wells in the sensor, so the signal to noise ratio is going to be way worse. With so much less light so you're seeing more of the read noise and dark current and all of that from the sensor that would normally be overrun with signal on a longer exposure. changing iso doesn't change the sensitivity of the sensor, but more light will give you better signal to noise performance. That's why when we do astrophotography we like to stack longer exposures whenever possible instead of just hundreds or thousands of really short exposures because the sensor has a noise floor and the longer you can expose, the better your S/N ratio is-- barring the sensor overheating and all of that... with anything there is a sweet spot you want to hit and that will vary depending on your tech.
@@FStoppers The only difference I see is that the DH1 uses a CCD sensor and the D750 & D850 use a CMOS sensor. CCD sensors create high-quality, low-noise images. CMOS sensors, traditionally, are more susceptible to noise. That is because each pixel on a CMOS sensor has several transistors located next to it, the light sensitivity of a CMOS chip tends to be lower. CMOS sensors are just now improving to the point where they reach near parity with CCD devices in some applications. Maybe why the test was not identical and you had different noise levels.
@@ronyedin i agree. i've read CCDs give better and cleaner image quality when i was searching budget camera bodies for astrophotography. but it's not as efficient and cost-effective as the widely popular CMOS sensor so camera manufacturers moved away from it.
raizen82 but a cleaner file shouldn’t mean the ISO standard is suddenly different because it’s cmos or ccd! That’s like saying their should be a different rating of ISO for Ilford and Kodak films. A sensitivity rating is a sensitivity rating. -P
@@FStoppers that's why i think your first test already showed ISO isn't actually a standard but arbitrary both across sensor tech and company manufacturer or even within camera models from the same manufacturer
Can you please do a Mythbuster video on when, and at what size- you REALLY notice and see a difference in PRINT QUALITY between different sensor sizes?
@@jacquesvroom Thats not true. Things can be truely identical. For example, we know that each single electron has an identical electric charge. We know that there is no deviation at all. Same thing goes for quarks for example. On that micro scale we also do have truely random things you wouldn't encounter in a macro scale.
You really need a test image that has a bigger dynamics range, something with dark shadows and spectral highlights that both have detail. That’s where snr processing will shine.
@@bassangler73 ugh. There is no international standards organization. It is the international organization for standardization. And they say directly on their website, and if you want to call them and talk to them yourself, they will tell you it is no abbreviation. For the simple fact that the acronym would change in different languages. IOS if it is in english. And different orders in different languages. So the founders decided to use the 3 letter word ISO, which comes from the greek word isos, meaning equal. And in all of the company's videos, they themselves, pronounce it eye-so. Which again, you can check for yourself if you cared to do some research. 🙂
@@bassangler73 www.iso.org/about-us.html They say it themselves. On their own website. Iso is not an acronym or abbreviation. Iso is a 3 letter word the founders used which comes from the greek word isos which means equal. The company says it themselves. Directly on their own website. Not some reference website. Their own website. There is no international standards organization. It is the international organization for standardization. And again. If you wanted to hear them directly, just call the company yourself and see how they answer their own phones. I will tell you what you wont hear them say. You wont hear them say I.S.O.
I responded to a comment below and I am going to express here again for anyone else interested in knowing in simple words why in digital cameras ISO setting is not equivalent to changing the exposure in post, although for some of the newer sensors the visible difference is small (i.e. D850). The signal flows at it follows: signal collection-> signal amplification (camera hardware ISO setting)->Analog to Digital conversion-> write to file-> edit and possibly signal amplification in raw editing software. Because the camera sets the iso amplification BEFORE reaching the AD conversion, there are 2 factors: 1) any noise picked up BETWEEN the amplifier and A/D converter is not amplified, but it will be if you boost the exposure in post, this is actually the major visible impact (more noise in the image) this can be a dramatic difference for non "ISO-invariant" sensors, 2) the A/D converter is used more effectively with hardware iso (in camera ISO setting) because the signal is not compressed in the low levels before A/D conversion, otherwise leaving the majority of the available higher levels empty (if you look at the very dark image histogram before getting boosted you will notice that it is dramatically shifted to the left leaving the majority of it empty). So , even if you have a 14bit A/D conversion, you are effectively using a lot fewer bits, maybe even only 8 bits, if you are underexposing a lot, making your image far more prone to quantization. I believe that is the reason of the artifacts noticed in the very dark areas in the video (6:15), it is like trying to get the details of clouds when they are almost white, because only very few discrete levels are describing the gradation, you start noticing artifacts due to quantization.
yes and no. YES ISO sensitivity is getting better as technology advances. NO because he compared a DX sensor to a Full Frame that is why there is a 1 stop difference. Bigger sensor gathers more light. he was testing ISO not exposure that is why Lee did not bother to properly expose the D750 and D850. he stayed at 1/30th of a second on all test. watch 2:03
@@Dylon1981 bigger sensor does not gather more light. Take a light meter and meter a scene. There is no setting for sensor size. Higher pixel pitch affects light gathering resulting in less noise, not different exposure. Crop sensor takes the same image as full frame, only cropped. If you take a crop sensor camera and full frame with the same pixel pitch, exposure will be the same. Crop the full frame image to the crop sensor image, size it to the crop sensor image and the amount of noise is the same.
Well, with my d850, I can pretty much duplicate the results you got but I'm not sure it really proves anything except that boosting exposure in camera is the same as boosting it in Lightroom. When I properly expose images at different iso values, lower iso values always give lower noise. Seems to me that this is just a demonstration of the wide dynamic range of the sensor. Since I usually shoot landscapes, I think I'll stick with low iso settings to get the best images.
The differences between the gain applied in camera vs the gain applied on your editing software comes down to the fact that two different routines are in use. And you're right to wonder if some pre-processing doesn't happen before the raw file is written. In fact, some brands absolutely pre-process a little before writing the raw. Look at the StarEater glitch Sony has. They could fix it, but unlike Fuji, Sony craps on their customers.
So to reiterate the camera companies are processing high iso raw images so that the chromatic noise is shifted to a grayscale destination because that mimics film grain of old.
Keep in mind that the software processing between light room and the in camera processor is different and results may vary. You could try Capture One or Affinity and still see difference in the way noise is processed. They are close and this means the camera is "Iso Invariant". Try the same on a Canon 5D mkIV and the results between the files would be huge, because Canon is not really ISO invariant. The most ISO Invariant camera to my knowledge is the Olympus EM5MKII
My initial thought is that Lightroom could also be a contender to the fact that it may add noise by gaining, like if we had a perfect post processing system there is potential for it to be the same, it’s hard to say whether it’s right or wrong either way.
Thanks, I was going to test it myself. It was a wild claim from Tony, not what I would have imagined. It makes me think it is OK to play it safe and underexpose in some cases where the lighting is being difficult, if you need the higher shutter speed to freeze the action.
Most if not all manufacturers apply noise reduction to their raw files at least at ISOs above a certain threshold. This is admitted by some manufacturer's and demonstrated by the reduction in resolution at higher ISO settings.
based on the cameras I have had, the upper limit is usually where the fudging starts. I would say ISO 3200 and up would be where they lose accuracy. I would also shoot at F4 over F2.8 as every lens's T-Stop will likely be the same a stop or so stopped down vs. wide open. If you used the exact same lens on all 3 then it shouldn't matter. I would like to see a wide range of brands ( whatever you have ) shot to whatever numbers a good light meter says and tested at base ISO, middle ISO, and then ISO6400 or 12,800 to see how accurate they all are at low, medium, and upper ISOs. Use F4 on F2.8 lenses across the board so hopefully the light transmitted is equal among all the brands/models.
For post processing I'll submit that you would want to use several other products (Capture One, ON1 etc.) to see what they do to the RAW files. The "dust" you see in the blacks may not appear with someone else's RAW converter. Very good follow up to Tony's video.
The simple explanation for the difference between using a high ISO setting in the camera vs pulling out the detail from the black in PS is that both processes will be starting with the same information but will have slightly different noise reduction / detail deduction algorithms. That’s all. The algorithm built into the camera will have the advantage of deeper knowledge of the source data just due to the fact that it’s written by the manufacturer.
There is something going on. The boost is applied before saving the raw file. Sensor gets the information and converts it to analog signal. Then this signal is transfered to booster and after that there is an A/D convertor and after it the data is converted to RAW. So when you are at native ISO the boost is 1 - no boost is applied but when you are at higher ISO the boost become 2,3 or something else (this is the multiplier). By boosting the signal right after the sensor you tend to get lower noise because electrical signals get more noise if they are weak and are processed by a lot of devices. That's why you can't achieve exactly the same result when boosting in post, because you recorded a noisier signal at native ISO and it is noisier because it is underexposed. When it is underexposed the signal to noise ratio (SNR) is lower so when you boost in post you are boosting the noise as well. While when you are boosting right after the sensor it is a lot easier to cut ot the noise because the level of it is much lower than the signal level
It seems like you possibly misunderstood the first test. He's speaking about this rating ISO as it pertains to noise and image brightness you did not compare the noise. good to see the community working to prove or disprove these types of things. Thanks.
I did a test, I found my Panasonic G85 and Nikon D5300 are pretty much the same exposure at ISO 200, 3200, 6400, and 12,800. I shot at F5.6 on the Nikon 10-20mm 4.5-5.6 and Panasonic 12-60mm 3.5-5.6. The ISO 200 was at 1/6 sec, ISO 3200 was at 1/100 sec, ISO 6400 was at 1/200 sec, and ISO 12,800 was at 1/320 sec. I pulled them up 4 at a time in Aperture and looked at the overall exposure for the 4, 2 of each at once. I compared the 1/320 sec images separate as they were more exposed than the rest. ( 1/400 was not as easy to set for both cameras in my quick test ) I shot the same subject with the same constant lighting all at one sitting. I also changed the aspect ration on the Panny G85 to 3:2 to match the Nikon. I don't have a light meter anymore so I am unable to tell if these settings are proper for the light in the scene but was able to compare them all side by side on an Apple 30" Cinema display. Oh, I shot both at 30mm- 35mm equivalent focal distance.( 15mm for G85, 20mm for D5300 ) I can't say every model will be like this, I am referencing these 2 model camera only. I was very surprised myself at this outcome. Anyway, if anyone else has a few different models/brands, maybe they can show their results. I used F5.6 to make sure to minimize the possibility of the lowest F stop not being accurate in light transmission. I compared overall exposure from 200-6400 separately than the ISO 12,800 since I used 1/320 sec instead of 1/400. A very interest result for me. I expected the Nikon to brighter as I thought the m4/3 might be fudging on the actual exposure above ISO 1600 to give better detail but at less ISO. I did not find that at all. I can't say if each ISO is actually accurate but I can say, they appear to be equal to the other camera brand, at least the 2 brand/models I tested. I posted this same response on Northrup's ISO video as well.
I think with the "ISO vs exposure gain in post" argument, the point is that your mileage may vary based on your camera. It makes sense to me that it's just different software-based algorithms at play - one algorithm (ISO) is by Nikon, happening within the camera, and the other is Adobe's, happening in Photoshop, or Lightroom, or whatever. As always, we should test our own cameras out and see which method performs better for what we want to produce!
In the second test you have to remember that the precesing it's going to be different in camera is the manufacturer who process the raw file on PS its the adobe processing, maybe that why you see different artefacts when you zoom the file . Very interesting episode.
It really varies with the camera body. I've owned both types of cameras and the ones that do not add gain at the chip level tend to have less noise at the higher isos.
I am so happy to see a very respectable way of challenging another's claim. I saw Tony's video and was very surprised when I learned ISO in digital cameras. I felt as though I was taken advantage of due to the fact that my decision to get a D810 was based on a couple of things, one of which was the base ISO of 64 . But over time I believe one learns what works with their camera and that any claims are meaningless.. Anyway, Kudos for the respect given, it comes back to you! at least with me..
“The meter in this camera is telling me I’m over exposed by one stop”. That’s because what you just tested was the internal light metering of the D1H relative to the internal light meter of the other cameras. To test ISO variation between them you need to take an incident light meter reading of the scene, and set all the cameras according to that. Shoot a gray card and measure the values in photoshop to see what the differences is between the exposures. You’d expect the D1H to vary from the newer cameras because the matrix meeting is based on image sampling data and that changes as Nikon tweaks it between cameras. If you shoot in spot metering instead you have to insure you’re measuring the exact same spot, and that the field of view of the spot meter is identical. In other words if one spot meter samples 1° more area than the other, you won’t get the same suggested exposure.
He used manual settings, the ISO may not be correct compared to a light meter but it would show if the ISO is equal among cameras. I did it with a Nikon D5300 and Panasonic G85 and found both to be the same exposure using the same manual setting on a constant light studio shot.
Up to iso 400 the camera does analog gain. Above ISO 400 the camera combines the analog gain with digital gain, multiplication. Pushing exposure in post is all digital gain.Therefore, the image taken at iso 3200 or 1600 etc that has the iso400 equivalent of analog gain in it has more unique colors present in it, since it was gained to iso 400 before A/d converter.
Ian Norman from Lonely Speck had the same theory a few years ago which he tested with the Sony cameras. It's a fascinating experiment which shows that camera manufacturers aren't very transparent. Good work.
so this is incorrect, you have to check what the native ISO on the camera is and then apply that test from there. I might be wrong but in a camera like that the native ISO is probably around 640, so what he just did is put the image through 2 processes by taking a shot at ISO 100 to then bump it up later. The first is to have the camera process the image internally from Native to ISO100, and then bringing into PS and bumping the image from the already processed shot up to iso 3200 which is probably why the image looks worse than the one taken at iso 3200. So summarising the first shot was processed from native -- to ISO 3200 in camera, and the second shot was processed from native -- to ISO 100 in camera and then post-processed from 100-- to 3200 on PS. Hope this helps
ISO is applied gain after taking the shot. Theoria Apophasis is talking about it for years! ISO is applied gain (after taking the shot). So, if I protect the highlights shooting RAW, I will 99% of the time be fine in Post.
Tony is pretty spot on for the most part. I would have liked to see the top 3 (Canon, Sony Alpha, Nikon) different cameras to be able to test the ISO performance across the gamut of Prosumer cameras. I would expect that there is little to no difference between Nikon Cameras - but to test Tony's theory i would have used 2-3 different cam manufacturers. Still a great Fs Video.
I appreciate the well done video testing out Tony's video in a mature respectful manner, unlike other UA-camrs that act like immature clowns when they disagree with Tony.
You know what: I think it doesn't matter much. This is because back in the day having "100" or "400" ISO it wasn't primarily for capturing CLARITY but for having DIRECTLY an end result that you want. Right now with digital software being able to adjust brightness/contrast/gamma: it's most probably irrelevant even if you post-process physical film after it's scanned so it's mainly SHUTTER-SPEED that still remains as something that will adjust clarity without a chance for post-processing to help much at that.
You're missing something. Some sensors do actually change at a certain iso. So my Sony A7iii switches to a different mode when you hit 640. It amplifies the sensors signal via analog gain which produces less noise. So the camera is isoless...as long as you're in the same "range". 500 vs 100 would be the same. 3200 vs 640 would be the same. But 3200 vs 100 would not be the same.
On the D850 too?
@@FStoppers Same. All Sony sensors. after 640, You reach the point of ISO-Invariant.
www.dslrbodies.com/cameras/current-nikon-dslr-reviews/nikon-d850-camera-review.html
This article mentions a dual iso. I'm just saying it's worth mentioning that it can be a variable. Then people can decide what they want to do with that info! Haha
All cameras are different but from the linked graph it appears the D850 has two levels of ISO invariance, one between 100-320 and one from 400 upwards (photonstophotos.net/Charts/RN_ADU.htm - and chose D850 on the right). Some cameras exhibit ISO invariant behaviour from 100 upwards, some like tha A7iii and apparently D850 have two levels of ISO invariant behaviour. Some like the 6D only exhibit ISO invariant behaviour after ISO3200. Probably the best write up about ISO invariance can be found on lonelyspeck or photographylife.
www.lonelyspeck.com/how-to-find-the-best-iso-for-astrophotography-dynamic-range-and-noise/
@@jeffluo9591 I'd like to know more about that, and I seem to be reading conflicting numbers here in the comments, while a quick search on the internet only yields articles with vague claims. You claim the thresshold is at 640, Alyn Wallace claims the threshold is at 400. Have any sources so I may know where the threshold actually is? (interested on the D750 and D850 values)
Tony gets criticized a lot for some of his opinions, but I’ve found that he’s usually able to back up his claims in pretty reliable ways.
As a digital media teacher whose classroom is right next door to a traditional film classroom, I can confirm that iso from film to digital is a mess of confusion.
Read my comment above, unless you have an "ISO-invariat" sensor Tony is very wrong. If you expect to see what he states, to say with the just announced Canon RP you would be very disappointed.
@@armandot9137 even with an ISO invariant sensor there is more bit depth in the highlights of a raw file than the shadows, if you're pulling information out of the shadows you'll lose some amount of detail compared to if you stored them in the highlights of the file by using a higher ISO. on the other hand using a higher ISO costs you dynamic range.
@@TimSheehan I absolutely agree. I just did not feel like needing to go in the deeper details ;-)
@@armandot9137 Regardless of the method used to increase and decrease the sensor sensitivity, ISO standards should yield the same results in exposure (within a margin of course).
Dude...Tony has no idea how to read a flash meter....Nor how it works. That channels is a joke.
Photographers only just finding out what ISO is and us astrophotographers just enjoying the popcorn
@@tobiasyoder it is indeed but we're far more scientific in our approach (and have a million more reasons to understand ISO properly). But we're kinda fed up of people saying false statements like "higher ISO = more noise". Nice to see people are waking up to ISO invariance though, it's highly useful knowledge
Lol hoping astrophotography sounds as cool as astrophysics. No different from food photography. Anyway, from what I've gathered, ISO is just gain applied after the exposure.. however the gain applied by the camera's amplifiers in the signal chain is different from the gain you can get from software. Maybe that's why the exposures in this video looked different.
@@Triple070007 yep it's gain but the difference is that some cameras induce noise into the image after the amplification gain in camera. This is why there's a difference with boosting in post with some cameras. Cameras that don't add much noise after the amplification in camera are those that behave iso invariant so you can boost in post and get a similar result
Some modes of applying exposure gain are better than others. People super into photography tech are still adorably unaware of how anything about cameras work. News at 11.
@@AlynWallace same boat here, seen too many time people doing astro saying "I kept ISO low to keep the noise low" while they actually destroyed the image.
Ironically enough an article was recently published on Fstoppers explaining ISO the wrong way :)
Thanks for keeping it civil and respectful two of my favorite channels, hopefully one day you guys will collab.
Johannes Lopez gotta get Tony to Puerto Rico first :) -P
@@FStoppers I am sure Tony would love to :-) love your channel guys and the website!
@@FStoppers Seeing this a month later but OK :)
@@TonyAndChelsea This would be so lit 😂😂😂 All 4 of you!
The D850 has a second gain boost at ISO400, so that’s probably why you didn’t get exactly the same result. Try comparing ISO400 to ISO3200.
Some of the noise comes from the rounding of the number. In a 14 bit raw file, the intensity of a pixel changes from 0 to 16384, so a correctly exposed picture's rounding error is almost negligible.
However, when you underexpose by 5 stops, you effectively have 8-9 bit raw file bit intensity varying from 0 to 200-500, where shadows of this image have a small number of intensity. Then the rounding error becomes significant.
Few years ago I’ve read a research paper about photo sensors. The skinny of it is: sensors do have constant dynamic range but photo cameras also have hardware noise reduction that cleans up RAW data before saving it. Some dynamic range is cut on the bright side in the process, that’s why you can’t recover overexposed high ISOs. Extended ISO range is when hardware NR is disabled or inactive so you basically get same results as Camera RAW exposure slider. And there is Red cinema cameras that function exactly as Tony described - if sensor itself wasn’t clipped and there is still signal in shadows then you can totally recover anything. However Red has new Gemini camera with dual ISO and it’s low light ISO setting does have hardware NR, so you get lesser dynamic range, but also much cleaner image with higher overall exposure
The D1H has a CCD sensor and all newer bodys have a CMOS Sensor.
That could affect your ISO -settings.
At a basic level of understanding, the ISO is just brightening the image; But for a better result, its better to do this before committing the data to file than after. A digital file format has limitations at extreme dark or light dynamic ranges, because the information is quantized in to a small part of the dynamic range. The result is that the file cant hold as much information from a dark image than a correctly metered image. This I think is the main reason for the differences.
You are most right about what this video couldn't figure out. Nature has almost unlimited colors but digitization must limit the color palette one way or another.
It's actually a form of compression artifacting but in a small scale since those files are technically(by name) uncompressed.
5:07 "OK, I'm going to do this live, on camera 5:11 :)
It's a lot like recording audio;
Good Level + no post gain = better quality and less distortion.
Low Level + post gain = worse quality, background noise, artifacts.
Musician?
I agree. I do sound recording/mixing. Getting a good level into your mics trumps all just like getting good light levels in photography. It means you can drive your preamps lower (equivalent to using low ISO). Analog gain circuitry like mic preamps adds noise especially if pushed too far. In that case i opt to stay a bit lower and use post digital gain later if necessary. It seems to keep things cleaner.
This will practically change how I go about my shoots. I used to be scared of underexposing, now I guess there's one less thing to worry about in the field. Of course we as photographers should do our own further testing with our own camera models. I'm gonna do that right away after this
I think you guys made the same point but maybe I'm confused a little. In your first test, if the D750 and D850 shows the photo is overexposed @ ISO 400, doesn't it mean that you can get the same exposure as the D1H @ a lower ISO?? Hence the selling point!
That was my impression as well. If they were exaggerating “ISO” performance, the image would appear brighter at the same ISO setting - which it did.
Yes, it should have been shot at 200, right?
Tony was arguing that you would take a photo with an old camera at iso 3200 and a new camera at iso 3200 and notice there is less noise in the new cameras image at the “same” iso. While in reality it might actually be shooting a darker image at iso 1600 but it changed the number to 3200
Something is missing here, the D1h is using CCD sensor while others are CMOS sensor. So I reckon that the ISO calculation maybe different because of this.
Try d700
This is one of the reasons I love UA-cam. The near real-time interaction and reaction to other UA-camrs.
Who says the Lightroom algoritm is identical to the ones in-camera? People tend to take LR as THE ONLY and CORRECT post processing tool, no matter how many times we have seen that LR does not handle a specific raw file from a specific camera any good ? Heck they seem to still use the dreaded recovery-tool before you even see your photo...
Very solid test. Me myself found low ISO 5-stop-pushed image has practically more noise than a high ISO one.
What Tony says if I understand it correctly is that if that the image is pushed 5 stops in the camera the images would look exactly the same?? All software process the images differently. Lightroom would not do it in the exact same way as the software in the camera.
Magnus Eriksson I want to try this but Nikon’s in camera NEF editor only pushes an image 2 stops. Maybe their proprietary NEF software ( is it still capture nx?) would do the same thing? -P
Tony just exposes his ignorance when he claims that all cameras exhibit "ISO invariance". ADC (analogue to digital conversion) is handled differently by different sensors. Some use analogue gain in which case Tony's claims fall flat. Then there are the modern dual gain sensors where ISO invariance holds but only in two separate ISO ranges. This stuff is really not complicated.
Bingo!
There is one important comment to Tony's video to pay attention to, that the ISO is an analog gain applied to the signal (which is done as the gain on a stereo, using voltages) before it is converted to digital. Once the signal is digital, you have lost the fine details, and boosting it in post will cause "quantization errors", which looks like noise.
Example: An analog signal can represent any value between, say, 0 and 1. It can be 0.1, 0.123 or whatever. A digital signal can only be either 0 or 1. So when the analog signal is digitized the values are rounded.
First of all.. ISO in sensor is obtained by changing the Vcc voltage of the sensor. Although it's not canged for every ISO setting rather is decided in bands, for eg. 100-400 have the same input voltage of the sensor and the rest is processed by the processor.
Second, the Exposure settings in LR or PS are complex algorithms simulating exposure equivalence but is not eqivalent.
Third, in the internal processing of an image some other things are applied, like dark current clipping, witch in Nikon cameras is pretty hard, and this is the reason untill d850 and partially d800a the Nikon cameras where not the one used for astrophotography.
Dark current clipping is omitting the darkes informations in the image, to be breaf.
Fourth, the RAW file is nothing like the the true image comming from the sensor, it is heavily processed, not only in sharpness or contrast... The real image from the sensor is mostly magenta in tone and you would never be able to match it to the scene by hand in PS.
It was. But is it now days on iso invariant sensors?
No. The sensor's sensitivity has *nothing* to do with its digital supply.
Analog gain is a thing, but you have no idea what you're talking about.
Multiple gain architecture is only used on some sensors, others only have one analog gain set in silicon. The real image from a sensor, when debayered, will look very green, not magenta.
@@Spirit532 I don't understand what you say about the 'real' image which would look green. The bayer matrix of the sensors has twice as much green points than red and blue, so what would be the real image?
What I understood that the human eye is extremely good at distinguishing between slightest differences of greens but not as good on other colours, thus sensors use more green elements than other colours. And of course since three colours channels are recorded and you want to keep the rectangular and repeating matrix one colour has to have more sensor elements than the two others.
@@FreakAzoiyd You've answered your own question.
The difference you found is actually the process of of raw processing the "High ISO set" image to shift the "Chromatic Noise" to "Luminance Noise" which is more pleasing to a photographer because it looks more acceptable and filmic the "silver hallide clumping" yumminess of film grain is not always bad. It's also a way of mitigating huge file size as the resulting colour noise would make the file incredibly huge the colour averaging of the noise make it more manageable to save.
Great videos by both of you guys! So, we don't and will probably never really know if the "just boosting gain" theory is exactly correct. Even if the gain theory is correct as to which Tony suggests, the way Adobe Camera RAW interprets boosting 5 stops to get to 3200 ISO compared to say a Nikon RAW file "Before" it becomes a NEF file COULD be different. Hince, why you are seeing more grain in the 100 ISO boost shot. Either way, this is super interesting stuff and I am glad there are people like you guys and Tony who make videos about these subjects!!! :D Also, using your REALLY old Nikon might be too far apart generation wise. The sensitivity of the first generation sensors could have just been really bad on those really off compared to what we have now. But that is hard to say. Would be cool to test with a D700/D3 VS a D610/D750 to D810 to D850.
Jim Boomer this is what I was saying off camera. I wanted to test it by boosting the 100 iso file in camera with Nikon’s NEF processor. Unfortunately it only boosts the EV by 2 stops. -P
It's not impossible to know, just test it :)
Exactly - I suspect the files are different only because the algorithms used to boost the exposure in the camera are different than those used by ACR. So not necessarily more noise, just different noise. I think we DO know that it's just boosting gain - we just don't know exactly how each camera does it. The sensor would have to physically change for the actual sensitivity of the photosites to change. I also think older cameras are definitely "less ISO-less" than newer cameras.
These are not "great videos" at all. Both authors don't seem to know the first thing about ISO on digital cameras. Look up "analogue gain" vs "digital gain". Look up "dual gain sensor". Look up "ISO invariance" and understand when it does *not* apply.
This video was really quite well done. I really liked how you were so humble and not trying to find fault with Tony or anyone else. You're a class act!
Looks like you proved BOTH of Tony’s points and perhaps were just confused on the first test. The newer cameras make the same exposure look “brighter” (hence fake better ISO performance). The second test is also proven, and the “noise” difference Lee points out is splitting hairs and easily attributable to differences in the gain calculations between Nikon in-camera vs Photoshop in-computer.
I think you have it backwards. If ISO 6400 was actually ISO 4000 in order to make ISO 6400 look better then ISO 400 would be actually ISO320 if the pattern stays consistent. What I have found with all my Nikons and m4/3 cameras is that ISO100-800 seems to be the same across most cameras and only when getting to ISO 1600 and up do the shenanigans start where it reads a higher ISO than what it actually is. If ISO 12,800 is really ISO 8000 then a shot at that setting would look pretty good on a FF camera when the previous model maybe had the same look at shows ISO 10,000 but was actually ISO 6000. My m4/3 looks great up to 3200 and still very good at ISO 6400. If I had a light meter I could see what the meter shows I need at that ISO to see if the camera is lying or not. When I tested high iso, I just took images at ISO 1600 and up in 1/2 ISO steps and was able to see where the quality was too low for my standards for whatever size print. I tried with my Nikon stuff and found the variation among the 3 cameras I had but I didn't test the same exact settings on the same studio lit item with m4/3 to directly compare the same lighting and see how it matched up ISO wise.I have a D5300 for the Nikon 10-20mm, I will have to try it against my G7 and G85 and see if they are all the same at the same high ISO.
Yeah and noise reduction filters may have been set differently in the RAW file after the picture was shot. Nikon Picture Control can even affect some of that noise. For example, higher contrast setting..? I saw Tony's video before I saw this one a day or so ago. I agree with you, it would seem having more brightness at lower ISOs would be an advantage. And so iso 100 is actually more like iso 50 like what Tony says... I think that he proved his claim.
I'm confused. The shot that was pushed five stops clearly shows much more chroma and luma noise plus the funky white dots.How does that prove anything except shooting at 3200 produces better pictures?
Very interesting. This just changed my whole world on ISO. Now for some testing on my end with Fuji.
For the iso invariant test, your software matters. Capture One will give you a better result. Also, you get more highlight details in return, when you do that kind of boost, that's how you can get the "Film look".
Jonas CaptureOne rules!
@@ARMAJOV Indeed, but Capture one does some things automatically if you don't disable it. You can get (nearly) the same in Photoshop, it is just not the same set of defaults. In the tests in this video I wonder if Nikon does some high-ISO noise reduction like Sony does. If this is not deactivated, of cause there is more noise in the low-ISO picture. Tony did not mention those details. (According to some camera tutorials I think he must be aware of this in general. Maybe he left it out intentionally.) But the claims are more or less correct. The picture is the same within the sensor, the rest is signal processing. (Just ignoring possible dark frames or combinations of more than one picture in camera.) Reducing this topic to the exposure setting only is a bit too primitive thought. Lee is really no technician. @Lee, if there is a setting like Sonys high-ISO noise reduction, please disable it and re-shoot your example.
@@ulrichsiebald144 that was kinda my thiughts also. However the iso 100 pic has a ton of noise. Just different noise. If noise reduction was on its pretty crappy.
Watching both yours and Tony's videos today was fascinating. You didn't so much "bust" Tony's claims, as essentially prove the same things. Sure, there are very subtle differences, but overall I think you guys both nailed the basic concepts. But, despite all the techy stuff, the thing that impressed me the most about your video was the clear respect and admiration that you gave Tony, even though your conclusions were slightly different. This is EXACTLY the kind of robust debate that makes it a win-win for all of us. Thank you so much for being firm, clear, and very gracious.
We need to use a scientific standard to sort this out and call out fowl play.
The lumens could be measured from glass to sensor and sensor readout directly but we would need an efficient method of transfer after that to judge final output, manufactures probably wouldn't like this plus the different software engines they are running will have a part to play as well.
Not too mention, the frequency range of all light gathered and the levels from within would also need standardisation. A bit like measuring headphone performance (actual studio quality, not prosumer, rich kid or gamer BS).
Fowl play like.. chickens?
If you look on DXOMark, they actually test sensitivity vs advertised ISO, and from what I recall, there's quite a bit of variance, but most cameras are somewhere around half a stop less sensitive than their nominal ISO values suggest.
Birds play?
If you understand how photo data is saved as a matrix, you understand why there is more "pepper" noise when you bring up the exposure for the ISO 100 image. Depending on how many bits of data are saved to each pixel can vastly change the results. High ISO settings automatically multiply the pure raw data before being saved as a 10-bit ir 12-bit. Amplifying the ISO 100 image multiplies the already "lossy" 10 or 12-bit value AFTER being saved as a 10 or 12-bit.
Hi Fstoppers .. I don't think you found evidence for "the opposite to be true"
I think what you discovered in your old vs new Nikon teest is actually the fundamental problem Tony Northrop refers to.
The sensor technology between the old and the newer cameras improved so much that the older camera needs to boost the signal (gain) significanly more than the newer ones. Increased gain generally leads to more noise. So the improved sensor technology allows you to shoot the same picture at a lower gain and therefore noise.
Regarding ISO number inflation, there might be some truth to Tony's observation, until there is a clear standard/reference it can and will be used to make the product look better. Bigger Numbers make it look better for sales... same as GHz on devices, HP on cars etc. ...
Regarding your second test
Yes there can be some "processing" going one before the image is stored ... you need to digitise the analog signal
Every pixel in the end has a limited capacity of storing different information. Lets assume there are only 1024 brightness levels
Correctly exposed you will used camera gain and all 1024 Levels, but when underexposing 5 stops you cram all the information in just 32 levels
Because the gain gets applied after the values are digitised it is possible that this causes artefacts
Analog | 5 Stop Gain || Digital Value | 5 Stop Gain
7.968 | 255 || 7 | 224
8 | 256 || 8 | 256
8.0312 | 257 || 8 | 256
While the Analog values are actually very close to each other the post digitised values can be the same or very far apart
In an ideal world where the RAW file is able to store infinitely/sufficiently precise, there is no difference between the two
For now, using ETTR you can use the information capacity within a RAW file as much as possible.
Also high information RAW formats today allow you to be off by 2-3 stops, without loosing too much informatinon
.. and as always ... your mileage may vary ...
For any given camera, ISO is metadata-only (so doesn't matter if you change it in camera or in post - also called "ISO invariant") for a given range or set of ranges - usually above a certain ISO. For example the 5D MarkIV is ISO invariant between ISO400 and anything above - but below that it is NOT invariant. the 5D Mark II was not ISO invariant until, if I recall correctly, well above ISO 1600. Red cameras are purely ISO invariant, as is the Arri Raw codec. You have to look up the model of camera you are using to find out the range or ranges of ISO invariance to know. Testing one camera only tells you about that one model, and even then you would have to test all ISOs to find the range(s) of invariance.
You should retest using the same tstop and lens across all cameras rather than going off the fstop. Usually fstops are off on most manufacturers, check dxo mark.
I used the exact same lens at 2.8 so it would t change between cameras
In video this is a bit more known because videocamera’s often use gain measured in decibels. Alister Chapman had some interesting content about that. In the end, amplifying the signal on the sensor level will probably always be better then in post but the difference is small. RED camera’s even use ISO/gain as a metadata setting, you can set the iso in post!
When it involves skin tone, though, I think that ISO-100-pushed-5-stop photo will show unnatural look. I would shoot at ISO 3200 when shooting people in this setting.
Don't "think".... TRY !
It's not hard and itakes what 3 minutes to do and then you would KNOW. My god people are so lazy today.
It would have been a more interesting test if it was only 2-3 stops underexposed. I bet they would have recovered almost perfectly in that case. The fact you can recover five stops and even have a somewhat usable image is simply mind blowing.
@@elvirredzepovic6898 funny thing I make one people test from his d810 which is supposed to be iso invariant at 400 iso. And he said yes it's more magenta if you boost 640 iso to 6400 iso. Of course this moron didn't set the same white balance. So sometime people Cannot test correctly
That thing about those white dots is really important. I use an entry level DSLR and I have encountered huge problems with underexposing my images by mistake. Once when I underexposed my images a lot and brought them back in Lightroom some of the areas which were black turned completely white while editing. Lightroom just went crazy. If I underexpose by one stop it's fine but when I try to push it further the editing part gets tricky at times. ISO might not seem like a big deal after all, you just need to get it around the correct value but when it comes to us - entry level camera users or crop-sensor users in general the difference between setting the correct ISO might be huge.
Tony failed to mention the sensor needs to be iso invariant you need to check the list of iso invarient cameras. Most Sony’s are. Canon isn’t fuji xtrans is. Also there are two groups. Iso base -400 then iso 640 to 6400. My Sony a7iii is invariant as are my fuji. But the 5dmk4 isn’t. I didn’t think the Nikon was either. I’ll agree that applying gain in post as opposed to in camera has its benefits for example retaining highlights and better colours. But the in camera gain seems better than Lightrooms. I did it and explained in my video a few month back. As far back as the fujifilm xt1 was invariant. Weird that canon isn’t. It’s ideal for timelaps shoots as you can underexpose and retain better highlights and lift the exposure instead of clipping when it got too bright for example 😊👍🏻 there are a few cameras on here that are and are not. Check it out 😊❤️📷
improvephotography.com/34818/iso-invariance/
Most nikon's with sony sensors are aswell. My d7100 is not. If i push 5 stops all i get is color noise, banding and a green cast
the 5Dmk4 is "partly ISO-invariant".
"The 5D Mark IV isn't entirely ISO-invariant: pushing an ISO 200 underexposed by 5 stops by 5 EV in post-processing yields slightly higher noise levels than a native ISO 6400 exposure. An ISO 100 exposure pushed 6 stops fares even worse. However, above ISO 400, the camera does, for the most part, exhibit ISO invariance, meaning that you could underexpose a traditional ISO 6400 exposure by 4 EV by shooting it at ISO 400 (while maintaining the shutter speed and aperture for ISO 6400), and then raise exposure 4 EV in post. This technique would afford you 4 EV of highlight headroom, with little to no noise cost, relative to shooting at ISO 6400."
www.dpreview.com/reviews/canon-eos-5d-mark-iv/11
Even newer cameras apply analog gain to the sensor signal before running it through the ADC when you turn the camera ISO up. To a point, analog gain before ADC will result in a cleaner image as a significant amount of the noise you see comes from the ADC and a larger signal into the ADC is a cleaner conversion. The larger signal can be either more light, or more analog gain. More light is cleaner than more analog gain, but more analog gain is cleaner (by a lot) than gaining the signal up after the ADC (digital gain).
In your test with the three cameras, isn't it the oldest camera that has the darkest picture at the same ISO as the newer cameras? This result then isn't "the opposite". You have to lower the ISO setting of the newer cameras making it appear that they get the same exposure at a lower ISO.
This is interesting. I'm an old film shooter, did a lot of very tight technical exposures. What I noticed when I started shooting with a Nikon D2X is that it matched my Minolta IV meter (incident) exposures and was itself accurate enough to use as a meter. I'd set the D2X on P, then transfer the settings to a 4x5 or 8x10 camera (adjusting for bellows of course) and still get a very accurate exposure. That was 10-14 years ago. I also started using my D2X for B&W film settings but there's a lot more latitude shooting negative. When I first started shooting jobs that were part 4x5 /8x10 film and digital DSLR, the settings I got from my light meter worked well in my DSLR. So I would say that through the Nikon D2X ISO matched the long standing criteria.
It doesn't surprise me that some companies might fudge this. I don't think Nikon or Canon would do this on the pro-level DSLR gear.
Your 3200 then 5 stops underexposed really surprises me. What is the range of a camera sensor? A typical film shot is if I remember has on average about 7 stops in range in the image. This doesn't mean you can shoot + or - 3.5 stops, this means that at the bottom end is almost black (DMax) and at the top end is almost pure white. A difference in 3/4 of a stop is noticeable, less is pretty much the same (in transparencies) three shots 1.5 stops different, even in a color negative film means two of them will be useable. If you expose 5 stops under? In transparency film that is pure black (if I had a reflection that was - 5 stops, I could see the reflection, but on film there would be no reflection). I'm curious as to what the actual range the sensors capture raw data, that's the real determining factor. Good video.
Tony was spot on in his video 🙂. And this video just made me think one of his pet peeves is when people say I. S. O. and not ISO haha
Yeah that drives me nuts. It's not an acronym!
Indeed. More precisely, according to Wikipedia:
>> ISO is not an acronym. The organization adopted ISO as its abbreviated name in reference to the Greek word isos (ίσος, meaning "equal")
But in the end, hearing it as being pronounced "I.S.O." here and there is really not a reason to go nuts ;-)
Go out for shooting, that's better for the health than to discuss endlessly the right pronounciation of the 3rd member of the exposure triangle. Oh wait, it's not even that; Ken W. would bash the sh** out of me if he read this technically false statement...
International Organization for Standardization?? I’ve thought this for years. And previous to that ASA was the American Standards Association..?
@@korm87 yes previously asa was used and that was an acronym. Iso is not an acronym though.
www.iso.org/about-us.html
Straight from the company's own website. They say it themselves the founders used the word iso taken from greek word isos meaning equal.
Jeezus you guys, talk about splitting hairs. The organization is in fact an acronym I.S.O... But in order to keep the rating acronym consistent across different countries, they came up with some silly excuse about the greek meaning of Isos... Again, literally just to make sure every country said their ACRONYM correctly.
Actually getting annoyed and correcting that level of minutia frankly makes you look dumber than just using the acronym everyone understands.
I love it when my favourite channels reply to each other and even challenge claims of others (in friendly manner). No matter who is right in the end, it puts out quite a lot of interesting information out and a few forth and back discussions generate more tests and claritications.
This is what photographers have become into. Arguing about nonsense. Very sad
Knowing how to handle ISO and knowing if we have a set standard of ISO or not is not nonsense
You are not even convinced of each other's explanations and arguing who is right or wrong. This type of info could be answered only by manufacturers. So this is all nonsense. Goodbye
I could be wrong about this, but I thought I had heard from an engineer I was talking to at NAB last year that ISO numbers for digital sensors are all relative to the sensor itself. In other words, a sensor will have a “native” ISO (usually around 400) that represents the raw data from the sensor with no manipulation, and all other numbers just indicate the factor that native ISO has been manipulated. For example, if the native ISO is 400, 800 would simply be whatever manipulation would be needed to increase exposure by one stop, 200 would be manipulated to decrease by a stop, etc. In that regard, it would make sense that a newer camera with a better sensor would capture more light at its native ISO, thus causing the image to be overexposed when using the same settings as an older camera. It would also explain the differences in results as every sensor would have its own native exposure sensitivity.
In regards to pushing ISO in post as opposed to on camera, I always thought most DSLRs do some low-level noise reduction at higher ISO settings? That would explain the difference in results when pushing an ISO 100 RAW image taken with the same settings as a properly-exposed photo shot at a higher ISO.
Anyway, just my $.02.
The lost dynamic range is a significant thing to keep in mind with modern cameras. As you increase your ISO, you start clipping highlights within the RAW file. Once that file is saved, and that pixel is registered as a full well, there is no longer any data to be recovered.
So, if you shoot a file at ISO-3200 to get detail into the deep shadows, you will lose, beyond recovery, the details in the highlights. The reverse is not true, as you have just demonstrated. Even 5 stops of push (Which is a damed lot!), does not cause any significant degradation of the file quality.
Hence, the old rule of "expose to the right" is defunct.
Thats not exactly true. Certain cameras actually gain highlights as you go up in ISOs. For example. The blackmagic pocket Cinema camera 4k actually has more stops in the highlights as you go up in ISOs, then it resets when you hit the 2nd circuit of the dual native ISO
It’s not entirely untrue though. It certainly is true for most modern cameras.
People just need to remember that some cameras are ISO invariant and others are not.
Exposure to the right is not defunct at all, it's just that changing ISO has no effect on exposure. Exposure is a function of scene luminescence, aperture ratio, and shutter speed.
Same comment I made on Tony’s. Since ISO is gain you can not isolate the result from signal to noise. It can vary greatly, just as in audio equipment with the quality of the amplification (hardware) and the case of the digital the, software.
What the heck......Ive been making videos for years that ISO is not part of exposure . Who copied who
Cool story lady
You copied Tony angry photographer
Ah poor old jealous ken...
Go away...
@@Lionheart2323 facts are not stories, son.
I have noticed that some people are using the term ISO Invariance to mean "mostly the same" as opposed to meaning "Identical". Mostly the same, as I understand it, means that that the change due to gain ( before the ADC) is similar to the change due to doing some math in software. Which is a good thing. I guess it just means the engineers making the hardware and firmware have done quite well at reducing noise where it can be reduced.
It's ISO, not ISO.
**puts on boxing gloves**
Its I.S.O. not eye so...its an abbreviation for International Standards Organization
@@bassangler73 its international organization for standardization. Iso is a 3 letter word the founders of the company used, which comes from the greek word isos, meaning equal. This information is all public knowledge and free for anyone who cares to take the time to go to the company's website and look for themselves. Or you could call them and hear them answer the phone amd see if they say i.s.o. or eye-so (spoiler alert, they will say eye-so). 🙂
Gemini is also pronounced "Geminee."
@@Triple070007 in what language?
@@Lionheart2323 www.reference.com/technology/iso-camera-4e486148d690dcb2
Boosting in post is often indistinguishable from raising ISO in camera for under exposed images. BUT of course blown highlights can never be recovered in post so dropping the ISO must do something before the raw file is baked.
I think you misunderstood the concept of iso invariance, maybe talk to an engineer before you make these vids?
Uggg,....all image sensors are pre-amped when they are converted to digital. That conversion is set by the camera company to be whatever "ISO" they want their "0db" to equal. (some cameras have TWO "0db's"...low 0db and high 0db) and Remember...all raw sensor data is saved at "0db"....ONLY. There is NO USER ADJUSTABLE GAIN APPLIED TO THAT RAW COLLECTION. This only happens whee the raw data is assembled into a .jpg. The raw data ALWAYS stays at native "0db". If you add +12db to your .jpg, than a "+12db" FLAG is added to the raw metadata. It's only a FLAG given to your RAW reader to apply a default +12db on RAW assembly. The RAW data in the file still exists at "0db"...always. Gain is a destructive process and that's why it's NEVER calculated into RAW's 0db status. This is why raw has the highlight recovery ability that it does....because you CANNOT CLIP IT USING CAMERA GAIN LIKE YOU CAN A JPG!
My theory for why there are differences is that the sensor data gets first multiplied by the gain setting and than is stored in the same way as it gets stored in JPEGs, as square roots due to the way human perception of light works.
ISO is not an acronym, but do they have to shout? 📣🙉🤷♂️
The amplification value of the chosen ISO is added before the digital conversion of the signal, and not afterwards, like changing the exposure value in Lightroom would. This decreases the noise in the image.
That video really intrigued me - glad you're following up, hows Puerto Rico living up for ya
Incredible so far
Very good addon video, wish that someday there would be a frotalk gathering with fstoppers, tony& Chelsea and may be Ken Rockwell too. It would be fun to see you all together
Tony used Lightroom, you used Photoshop. LR has noise (and color noise) reduction sliders which if you have used them, then the image would have been exactly like the one taken from Nikon at ISO 3200. The camera simply applies some noise reduction techniques when shooting at an ISO higher than 100, in rest is just software gain.
Critical Point But does it apply noise reduction before writing the NEF file? I was always told raw files don’t have any noise reduction/sharpening/shadow recovery etc. -P
Critical Point - Then doesn't that prove his point, when treating both files the same way (meaning, not doing color noise reduction) doesn't that reflex that the files are not identical?
@@FStoppers Check out some of the forums on dpreview or photonstophotos, Nikon is known for baking in a level of noise reduction in blacks in the raw file itself.
Yes agree, I think RAW files are more pre baked. Its not RAW RAW without anything on it
Critical Point exactly! Different gain and noise reduction algorithms and calculations in camera vs software.
Remember with film when you pushed the exposure (2 stops at most) you got more grain. When you push it digitally you get more noise.
To be fair to Tony. Nikon is not exaggerating but Fuji absolutely exaggerates their claims
Really well done! This strikes me as a good example of how the scientific method should work. Tony posited some hypotheses, you guys test and find your results, then the data are what they are. I really appreciate that over stupid, drama-based flame wars.
Oh yes, one or two tests provides scientific proof now days. Fake science.
Oh here we go LOL
The iso 100 exposure's much shorter shutter speed will mean a lot fewer photons hitting the light wells in the sensor, so the signal to noise ratio is going to be way worse. With so much less light so you're seeing more of the read noise and dark current and all of that from the sensor that would normally be overrun with signal on a longer exposure.
changing iso doesn't change the sensitivity of the sensor, but more light will give you better signal to noise performance. That's why when we do astrophotography we like to stack longer exposures whenever possible instead of just hundreds or thousands of really short exposures because the sensor has a noise floor and the longer you can expose, the better your S/N ratio is-- barring the sensor overheating and all of that... with anything there is a sweet spot you want to hit and that will vary depending on your tech.
The reason the D1H and the D750/850 have different ISO performance is the D1H is APS-C.. Tony addressed this in his video.
Why would sensor size effect the iso number? Iso is suppose to be uniform across all cameras and sensor sizes.
@@FStoppers The only difference I see is that the DH1 uses a CCD sensor and the D750 & D850 use a CMOS sensor. CCD sensors create high-quality, low-noise images. CMOS sensors, traditionally, are more susceptible to noise. That is because each pixel on a CMOS sensor has several transistors located next to it, the light sensitivity of a CMOS chip tends to be lower. CMOS sensors are just now improving to the point where they reach near parity with CCD devices in some applications. Maybe why the test was not identical and you had different noise levels.
@@ronyedin i agree. i've read CCDs give better and cleaner image quality when i was searching budget camera bodies for astrophotography. but it's not as efficient and cost-effective as the widely popular CMOS sensor so camera manufacturers moved away from it.
raizen82 but a cleaner file shouldn’t mean the ISO standard is suddenly different because it’s cmos or ccd! That’s like saying their should be a different rating of ISO for Ilford and Kodak films. A sensitivity rating is a sensitivity rating. -P
@@FStoppers that's why i think your first test already showed ISO isn't actually a standard but arbitrary both across sensor tech and company manufacturer or even within camera models from the same manufacturer
Can you please do a Mythbuster video on when, and at what size- you REALLY notice and see a difference in PRINT QUALITY between different sensor sizes?
'almost identical' either it's identical, or is not. :)
It's not identical. That much is clear.
@@jidrztgc318 but almost
Kids, nothing is "identical"...
@@jacquesvroom Thats not true. Things can be truely identical. For example, we know that each single electron has an identical electric charge. We know that there is no deviation at all. Same thing goes for quarks for example.
On that micro scale we also do have truely random things you wouldn't encounter in a macro scale.
@@jort93z Do those "identical" charges have "identical" locations? :)
You really need a test image that has a bigger dynamics range, something with dark shadows and spectral highlights that both have detail. That’s where snr processing will shine.
ISO is not an acronym. It is a 3 letter word. 🙂
No its an abbreviation for International Standards Organization
@@bassangler73 ugh. There is no international standards organization. It is the international organization for standardization. And they say directly on their website, and if you want to call them and talk to them yourself, they will tell you it is no abbreviation. For the simple fact that the acronym would change in different languages. IOS if it is in english. And different orders in different languages. So the founders decided to use the 3 letter word ISO, which comes from the greek word isos, meaning equal. And in all of the company's videos, they themselves, pronounce it eye-so. Which again, you can check for yourself if you cared to do some research. 🙂
@@Lionheart2323 www.reference.com/technology/iso-camera-4e486148d690dcb2
@@bassangler73 www.iso.org/about-us.html
They say it themselves. On their own website. Iso is not an acronym or abbreviation. Iso is a 3 letter word the founders used which comes from the greek word isos which means equal. The company says it themselves. Directly on their own website. Not some reference website. Their own website. There is no international standards organization. It is the international organization for standardization. And again. If you wanted to hear them directly, just call the company yourself and see how they answer their own phones. I will tell you what you wont hear them say. You wont hear them say I.S.O.
@@bassangler73 Wrong. It's a word.
I responded to a comment below and I am going to express here again for anyone else interested in knowing in simple words why in digital cameras ISO setting is not equivalent to changing the exposure in post, although for some of the newer sensors the visible difference is small (i.e. D850). The signal flows at it follows: signal collection-> signal amplification (camera hardware ISO setting)->Analog to Digital conversion-> write to file-> edit and possibly signal amplification in raw editing software. Because the camera sets the iso amplification BEFORE reaching the AD conversion, there are 2 factors: 1) any noise picked up BETWEEN the amplifier and A/D converter is not amplified, but it will be if you boost the exposure in post, this is actually the major visible impact (more noise in the image) this can be a dramatic difference for non "ISO-invariant" sensors, 2) the A/D converter is used more effectively with hardware iso (in camera ISO setting) because the signal is not compressed in the low levels before A/D conversion, otherwise leaving the majority of the available higher levels empty (if you look at the very dark image histogram before getting boosted you will notice that it is dramatically shifted to the left leaving the majority of it empty). So , even if you have a 14bit A/D conversion, you are effectively using a lot fewer bits, maybe even only 8 bits, if you are underexposing a lot, making your image far more prone to quantization. I believe that is the reason of the artifacts noticed in the very dark areas in the video (6:15), it is like trying to get the details of clouds when they are almost white, because only very few discrete levels are describing the gradation, you start noticing artifacts due to quantization.
ISO getting better , that’s why 750 and 850 said over exposed. ISO has no universal standard.
ISO is a international Standard.
yes and no. YES ISO sensitivity is getting better as technology advances. NO because he compared a DX sensor to a Full Frame that is why there is a 1 stop difference. Bigger sensor gathers more light. he was testing ISO not exposure that is why Lee did not bother to properly expose the D750 and D850. he stayed at 1/30th of a second on all test. watch 2:03
@@Dylon1981 bigger sensor does not gather more light. Take a light meter and meter a scene. There is no setting for sensor size. Higher pixel pitch affects light gathering resulting in less noise, not different exposure. Crop sensor takes the same image as full frame, only cropped. If you take a crop sensor camera and full frame with the same pixel pitch, exposure will be the same. Crop the full frame image to the crop sensor image, size it to the crop sensor image and the amount of noise is the same.
Well, with my d850, I can pretty much duplicate the results you got but I'm not sure it really proves anything except that boosting exposure in camera is the same as boosting it in Lightroom. When I properly expose images at different iso values, lower iso values always give lower noise. Seems to me that this is just a demonstration of the wide dynamic range of the sensor. Since I usually shoot landscapes, I think I'll stick with low iso settings to get the best images.
The differences between the gain applied in camera vs the gain applied on your editing software comes down to the fact that two different routines are in use. And you're right to wonder if some pre-processing doesn't happen before the raw file is written. In fact, some brands absolutely pre-process a little before writing the raw. Look at the StarEater glitch Sony has. They could fix it, but unlike Fuji, Sony craps on their customers.
So to reiterate the camera companies are processing high iso raw images so that the chromatic noise is shifted to a grayscale destination because that mimics film grain of old.
Keep in mind that the software processing between light room and the in camera processor is different and results may vary. You could try Capture One or Affinity and still see difference in the way noise is processed. They are close and this means the camera is "Iso Invariant". Try the same on a Canon 5D mkIV and the results between the files would be huge, because Canon is not really ISO invariant. The most ISO Invariant camera to my knowledge is the Olympus EM5MKII
My initial thought is that Lightroom could also be a contender to the fact that it may add noise by gaining, like if we had a perfect post processing system there is potential for it to be the same, it’s hard to say whether it’s right or wrong either way.
Makes sense, not every sensor is the same, Sekonic pointed this out with their light meter profiles.
Thanks, I was going to test it myself. It was a wild claim from Tony, not what I would have imagined. It makes me think it is OK to play it safe and underexpose in some cases where the lighting is being difficult, if you need the higher shutter speed to freeze the action.
Most if not all manufacturers apply noise reduction to their raw files at least at ISOs above a certain threshold. This is admitted by some manufacturer's and demonstrated by the reduction in resolution at higher ISO settings.
based on the cameras I have had, the upper limit is usually where the fudging starts. I would say ISO 3200 and up would be where they lose accuracy. I would also shoot at F4 over F2.8 as every lens's T-Stop will likely be the same a stop or so stopped down vs. wide open. If you used the exact same lens on all 3 then it shouldn't matter. I would like to see a wide range of brands ( whatever you have ) shot to whatever numbers a good light meter says and tested at base ISO, middle ISO, and then ISO6400 or 12,800 to see how accurate they all are at low, medium, and upper ISOs. Use F4 on F2.8 lenses across the board so hopefully the light transmitted is equal among all the brands/models.
For post processing I'll submit that you would want to use several other products (Capture One, ON1 etc.) to see what they do to the RAW files. The "dust" you see in the blacks may not appear with someone else's RAW converter. Very good follow up to Tony's video.
The simple explanation for the difference between using a high ISO setting in the camera vs pulling out the detail from the black in PS is that both processes will be starting with the same information but will have slightly different noise reduction / detail deduction algorithms. That’s all. The algorithm built into the camera will have the advantage of deeper knowledge of the source data just due to the fact that it’s written by the manufacturer.
There is something going on. The boost is applied before saving the raw file. Sensor gets the information and converts it to analog signal. Then this signal is transfered to booster and after that there is an A/D convertor and after it the data is converted to RAW. So when you are at native ISO the boost is 1 - no boost is applied but when you are at higher ISO the boost become 2,3 or something else (this is the multiplier). By boosting the signal right after the sensor you tend to get lower noise because electrical signals get more noise if they are weak and are processed by a lot of devices. That's why you can't achieve exactly the same result when boosting in post, because you recorded a noisier signal at native ISO and it is noisier because it is underexposed. When it is underexposed the signal to noise ratio (SNR) is lower so when you boost in post you are boosting the noise as well. While when you are boosting right after the sensor it is a lot easier to cut ot the noise because the level of it is much lower than the signal level
It seems like you possibly misunderstood the first test. He's speaking about this rating ISO as it pertains to noise and image brightness you did not compare the noise. good to see the community working to prove or disprove these types of things. Thanks.
As far as grain goes, I would expect that the camera includes a noise pass, not just a gain boost
I appreciate adult discussions between professionals, thank you fstoppers another wonderful video sans any “fro” drama.
I did a test, I found my Panasonic G85 and Nikon D5300 are pretty much the same exposure at ISO 200, 3200, 6400, and 12,800. I shot at F5.6 on the Nikon 10-20mm 4.5-5.6 and Panasonic 12-60mm 3.5-5.6. The ISO 200 was at 1/6 sec, ISO 3200 was at 1/100 sec, ISO 6400 was at 1/200 sec, and ISO 12,800 was at 1/320 sec. I pulled them up 4 at a time in Aperture and looked at the overall exposure for the 4, 2 of each at once. I compared the 1/320 sec images separate as they were more exposed than the rest. ( 1/400 was not as easy to set for both cameras in my quick test ) I shot the same subject with the same constant lighting all at one sitting. I also changed the aspect ration on the Panny G85 to 3:2 to match the Nikon. I don't have a light meter anymore so I am unable to tell if these settings are proper for the light in the scene but was able to compare them all side by side on an Apple 30" Cinema display. Oh, I shot both at 30mm- 35mm equivalent focal distance.( 15mm for G85, 20mm for D5300 )
I can't say every model will be like this, I am referencing these 2 model camera only. I was very surprised myself at this outcome. Anyway, if anyone else has a few different models/brands, maybe they can show their results. I used F5.6 to make sure to minimize the possibility of the lowest F stop not being accurate in light transmission. I compared overall exposure from 200-6400 separately than the ISO 12,800 since I used 1/320 sec instead of 1/400. A very interest result for me. I expected the Nikon to brighter as I thought the m4/3 might be fudging on the actual exposure above ISO 1600 to give better detail but at less ISO. I did not find that at all. I can't say if each ISO is actually accurate but I can say, they appear to be equal to the other camera brand, at least the 2 brand/models I tested. I posted this same response on Northrup's ISO video as well.
I think with the "ISO vs exposure gain in post" argument, the point is that your mileage may vary based on your camera. It makes sense to me that it's just different software-based algorithms at play - one algorithm (ISO) is by Nikon, happening within the camera, and the other is Adobe's, happening in Photoshop, or Lightroom, or whatever. As always, we should test our own cameras out and see which method performs better for what we want to produce!
Increasing the iso in camera boost's the analogue signal before it has been digitised and processed, so it will be better.
In the second test you have to remember that the precesing it's going to be different in camera is the manufacturer who process the raw file on PS its the adobe processing, maybe that why you see different artefacts when you zoom the file . Very interesting episode.
It really varies with the camera body. I've owned both types of cameras and the ones that do not add gain at the chip level tend to have less noise at the higher isos.
I am so happy to see a very respectable way of challenging another's claim. I saw Tony's video and was very surprised when I learned ISO in digital cameras. I felt as though I was taken advantage of due to the fact that my decision to get a D810 was based on a couple of things, one of which was the base ISO of 64 . But over time I believe one learns what works with their camera and that any claims are meaningless.. Anyway, Kudos for the respect given, it comes back to you! at least with me..
“The meter in this camera is telling me I’m over exposed by one stop”. That’s because what you just tested was the internal light metering of the D1H relative to the internal light meter of the other cameras. To test ISO variation between them you need to take an incident light meter reading of the scene, and set all the cameras according to that. Shoot a gray card and measure the values in photoshop to see what the differences is between the exposures.
You’d expect the D1H to vary from the newer cameras because the matrix meeting is based on image sampling data and that changes as Nikon tweaks it between cameras. If you shoot in spot metering instead you have to insure you’re measuring the exact same spot, and that the field of view of the spot meter is identical. In other words if one spot meter samples 1° more area than the other, you won’t get the same suggested exposure.
He used manual settings, the ISO may not be correct compared to a light meter but it would show if the ISO is equal among cameras. I did it with a Nikon D5300 and Panasonic G85 and found both to be the same exposure using the same manual setting on a constant light studio shot.
A high gain setting with a CMOS sensor results in lower readout noise, which I am sure accounts for the noisier low gain (ISO) image pushed in post.
and just to expand, readout noise is absolutely negligible under normal lighting conditions, only becoming apparent in low light
I read an article that claimed that even on the ISO invariant sensors, you lose dynamic range if you don’t set the exposure properly.
Up to iso 400 the camera does analog gain. Above ISO 400 the camera combines the analog gain with digital gain, multiplication. Pushing exposure in post is all digital gain.Therefore, the image taken at iso 3200 or 1600 etc that has the iso400 equivalent of analog gain in it has more unique colors present in it, since it was gained to iso 400 before A/d converter.
I think Tony meant that you have to shoot at the sensor's base ISO to boost it and get that identical photograph. That depends on the camera.
Ian Norman from Lonely Speck had the same theory a few years ago which he tested with the Sony cameras. It's a fascinating experiment which shows that camera manufacturers aren't very transparent. Good work.
You can basically only shoot at your native (analog signal) ISO, because everything else is a digital (software) adjustment.
so this is incorrect, you have to check what the native ISO on the camera is and then apply that test from there. I might be wrong but in a camera like that the native ISO is probably around 640, so what he just did is put the image through 2 processes by taking a shot at ISO 100 to then bump it up later. The first is to have the camera process the image internally from Native to ISO100, and then bringing into PS and bumping the image from the already processed shot up to iso 3200 which is probably why the image looks worse than the one taken at iso 3200. So summarising the first shot was processed from native -- to ISO 3200 in camera, and the second shot was processed from native -- to ISO 100 in camera and then post-processed from 100-- to 3200 on PS. Hope this helps
ISO is applied gain after taking the shot.
Theoria Apophasis is talking about it for years!
ISO is applied gain (after taking the shot).
So, if I protect the highlights shooting RAW, I will 99% of the time be fine in Post.
Tony is pretty spot on for the most part. I would have liked to see the top 3 (Canon, Sony Alpha, Nikon) different cameras to be able to test the ISO performance across the gamut of Prosumer cameras. I would expect that there is little to no difference between Nikon Cameras - but to test Tony's theory i would have used 2-3 different cam manufacturers.
Still a great Fs Video.
I appreciate the well done video testing out Tony's video in a mature respectful manner, unlike other UA-camrs that act like immature clowns when they disagree with Tony.
You know what: I think it doesn't matter much. This is because back in the day having "100" or "400" ISO it wasn't primarily for capturing CLARITY but for having DIRECTLY an end result that you want. Right now with digital software being able to adjust brightness/contrast/gamma: it's most probably irrelevant even if you post-process physical film after it's scanned so it's mainly SHUTTER-SPEED that still remains as something that will adjust clarity without a chance for post-processing to help much at that.