You are awesome, Michael, I love your content, you are mostly the reason I got into photography and really helped me chose my cameras and lenses... I wish you a happy new year and may your channel grow 100 times in 2020!

Great explanation Michael, should be required viewing 😄

Happy New Year Michael!!! Love the channel, keep up the good work.

Excellent analysis Michael and very well explained.

You never cease to deliver accurate information in a clear manner. Wish more of the KZfaqrs would cater to more "objective educating" as you do, and do less subjective shilling (you know the ones claiming "sold off all my X to switch to Y" or "DSLRs are dead and if you're not on mirrorless, you're behind..." ;-)

Very helpful. Thanks.

Wow ❤❤❤❤. Thank you so much.

Well, correct me if I am wrong but the way I see it is that the DXO P-MPIX score is in mega pixels in relation to the original sensor. So, in your example, the 24mm lens resolves 65% of the 20mp sensor of the D500 (which is close to 13mpix) while it resolves about 63-64% of the sensor of the D810 which is around 23mpix, so, in summary, the lens has about 63-65% resolving power, no?

Outstanding video. This issue of image sharpness and use of FF lenses on crop bodies has been exploited by bad information on the internet. You have brought clarity to the discussion. Please educate Mr. Northrop.

Michael my are only taking blackout photo. I went in to all the setting and it is still doing the same Thing over .yes it is canon t6s can u give some advice.

Sounds like they used the "fudge factor." Good review, Mike!

awesome! love your videos with such great technical explanations! Please make a video about flange distance, mount diameters, angle of incidence, etc of current mirrorless systems; I think there is still confusion around and I just dont get why a shorter flange distance should improve image quality. "Greater diameter mounts allow more room for lens design", but how can it be since the rear part of the lens has always been smaller than the sensor anyway?

Awesome education...occasionally we village people need to be reminded of the excellence that people such as your self strive to obtain when presenting your knowledge of the subject matter at hand...whew...way to much verbiage... I just love to hear you explain things...Back home on the mother world you are spoken of with great respect... as always keep up the good work and keep'em flying... I crack myself up...

It would be ideal of DXO had a way of testing every lens on equipment that could outresolve ANY lens no matter how sharp then simply give that metric as the sharpness score without combining it with amount of coverage the lens provides (apsc- full frame, MF). Something not mentioned here is how increased pixel density can cause diffraction to occur at lower f stops. Short story, early on in my digital career I purchased a Canon D60 (yes, they put the D first then), it was a 6 megapixel camera with 1.6x crop factor. My cousin on the other hand being a professional bought the 1D which had only a 4 megapixel sensor. We often shared lenses and his always seemed sharper but moreso when it was on his camera than mine, erroneously we came to the conclusion that the pixels in his pro camera were somehow higher quality but in hindsight it was likely that none of our lenses (canon l series v1 and some tamron) just didn't have great resolving power and the combination of using less of the image circle created by the lens and higher pixel density was simply showing inadequacies of the lenses. Personally I like reviews from other youtubers like Dustin Abbott and Christopher Frost, they show real world examples of lenses and compare them at different apertures on both APS-C and FF cameras. Kudos to yourself and others that make these extremely helpful videos for use to learn from.

Take that Tony!!! :-p

This is excellent content right here Michael!!! My background is in Philosophy and while I understand your point of view in not wanting to engage semantics I have to add from my own perspective that the phraseology of "percieved " megapixel count is inherently flawed and lends itself to circular tautology. In point of fact it "renders" their data as given as unscientific by definition. It's admirable yes, that they're trying to help the consumer - but it's equally admirable for you to take this position. Bravo!!

i currently consider: - canon ef 100mm f/2,8L macro is usm - sigma 85mm f/1,4 dg hsm art for canon eos 6d mark ii. i referred to dxomark and found 100mm has the same dxomark score with my 50mm f/1,8 stm while 85mm has 9 scores more than those two. the 100mm is basically two lenses as one - portrait and macro lens, and 85mm is portrait only. i decided to buy the 100mm one since there are clearance sale and that lens are on great discount while 85mm isn’t. is the 85mm one better for my overall image? i see 50mm has images that is _crazy_ sharp for me, but 85mm one ‘teases’ me with that score. man, picking lens is hard!

Michael - Thank you so much for this video! This is a topic I've been very interested in for years. Tony Northrup has been saying for years that you shouldn't use full frame (FF) lenses on crop body cameras and references DXO as support for his argument. This never made much sense to me since when using a FF lens on a crop sensor camera, the crop area is from the best performing part of the FF lens. Granted you may be paying more than necessary and dragging around a bigger lens because it was designed for a larger sensor But the performance is just as good, or even better on a crop sensor vs FF sensor since the imaging circle is the sweet spot/center area on the crop sensor. Thanks again for your excellent analysis!

I wonder what John Sheehy would say about this.. I"m going to post the link to one of the DPR forums.. ;) Thanks for all you do. Your delivery is very clear and logical (and thanks for providing a bit of info about your qualifications.).. CD Mazoff (PhD, McGill)

I've heard some in the industry criticize the DXO methodology before, but this really lays it out. Would be helpful though if they figured out how to test Fuji X-Trans sensors and glass at some point...

Wow! Excellent video. My results bare this out as well! I shoot wildlife and aircraft. The results of my full frame lenses on my D7200 are noticeably sharper than my D750. The D750 has an anti-aliasing filter and a courser pixel pitch. They are both 24 mega pixel cameras.

Thanks Michael, valuable information. Heard about DXO before (whenever a new camera phone releases) but did not know how to interpret it. Just looking at it now, although could not find 90D but noticed that the EFS 10-18mm shows low DXO score for almost all crop sensors (for e.g. checking for 7D Mk II, the 24mm f/2.8 has a score of 20 and the 10-18mm has a score of 12. Understand that a prime lens could have a better score in general, but wondering if there is anything else at play here? From the lenses discussion earlier, it was established that 10-18mm is a good lens for the 90D.

Michael, advanced microscopy PhD here (via materials science). You are absolutely in the right mindset and I would further add that DXO should forget about P-Mpix and rather focus on a pixel pitch-relatable metric instead (i.e. the resolving power of the lens measured at the plane of the image sensor) in their reviews, taking sensor size out of the equation. In fact, pixel pitch and the optical resolving power of the lens is what we as the photography community should understand better in order to avoid the ever so outdated debate about FX vs DX. Of course only with respect to sharpness, there are tons of other things that make a lens appealing to us photographers. DXO - do - however inform most often that a lens has the resolving power of XX um, relatable to pixel pitch. Unfortunately, DXO then make things much more intricate (in my view) by introducing P-Mpix and this metric will of course vary depending on whether we are talking about, say. micro-4/3, DX, FX or medium format. Add to that the black-box "penalty" and, wow, what a can of worms. Thanks, and keep sharing! Just found your channel so you have a new subscriber!

Even though we dont know exactly the methodology DXO use to arrive at their scores, if they are using the same methodology on all the different gear Ive always figured the results would still give us proportional points of reference to go off. Your thoughts?

It's an interesting topic to consider. And you are right: it is misleading of DXOMark to use the term "sharpness" when what they are really reporting is "resolution" for a given lens/sensor combination. I think a counter-argument might run this way: the Nikon D500 uses a 20.9MP sensor ; whereas the D850 in DX mode (as you did here) uses 19.4MP of its 46MP sensor - a near-identical match. The lens performed similarly and not, as DXOMark's "sharpness" rating reports, almost half as well. However, had you used the D850 in FF mode, the totality of the output from that same lens in terms of resolution across the D850's full-frame sensor would appear to accord with DXOMark's misleading "sharpness" metric. And that's not to mention DXOMark's testing practice of averaging this dodgy metric across the entire range of a given lens's aperture and focal lengths (in the case of zooms). By then, the number becomes not at all useful to prospective purchasers. DXOMark is a French company, so maybe something about the word "sharpness" got lost in translation at an earlier stage in their evolution, and they were disinclined to change it subsequently.

Finally !! Someone gets it! Now please next do a video that discusses why crop sensors are not always “worse in low light than FF”. It’s not the crop, it’s the pixel size!! Thanks Michael keep it up pls.

I am under the weather and not on the top of my game, but with sensors we are not talking linear but squared. The power of 2 and or square root should be in the calculation, at least for the area of the sensors that do not match up exactly. The larger defect will always be I bought the lens so it must be sharp bias.

Tony Northrup debunked in 3...2...1. Lmbo! This video is so good Michael. Thank you for doing this. I genuinely was fooled and was about to pull the trigger on a D810 because it had better sharpness than my D500. My 300mm f2.8 AFS ii is plenty good. No sense in spending that money since I actually NEED the extra reach. This upgrade would make sense if I had the 500mm f4 and could fill the frame but I only have a 300mm f2.8 and fill the frame on my crop sensor camera. Basically to see any upgrade I would just have to get closer.

One thing you need to add is when the image is observed and what is involved in that process. My background is both in photography (since the 60s and 70s when I did darkroom work for the zoology Dept at the University of Texas while in school) and high end offset lithography and digital prepress which ended up being my career of 33 years. The factors involved are the magnification from sensor image to final observed image, rasterization resolution capability of the output device, and the ability (or lack thereof) of the media to resolve the rasterization into a viewable image (soft surfaces don't resolve the detail, hard surfaces can bleed the inks about) and the distortion of what is being observed on just about any LED viewing screen. Scale of this parameter - screen I'm using has a need to max in out of 153K/ sq in of data to maximize what it can show, typical high end ink jet is capable when tweaked out of theoretically imaging at a max of 2.88 Mpx/ sq (600x1200 in of viewing area which would be able to resolve more from the lens/sensor than my screen by a lot. The Rampage print rasterization system's highest resolution would need 51.84 (3600x3600) to maximize it’s capabilities but that is both impractical and uneconomic and our best case situation reduced that to 360K for offset dot frequency best case (300 dpi). Typical was 2400x2400 to keep pixelation out of vector elements. To resolve tiny detail (magazine covers etc) our Hell Drum scanners wanted nothing less than medium format slides which they scanned at some ungodly high rate (10K to 11K dpi, 120 megs per sq in) which were then downsampled into the final output size and frequency to match the job quality requirements (press capability, pocketbook capabilities, etc.) This all only has applicability to your presentation in that when someone says a lens isn't sharp, I immediately ask what output format/process were you evaluating and at what magnification. This human side of the equation I think is just as important because of the amorphous notions involved in lens evaluations such as "acceptable circle of confusion". I like what you're providing the community.

Although I don't pay too much attention to DXO's metrics I enjoyed this presentation. Maybe you've done this already but I would be interested to see a full explanation about, or a debunking of, the commonly promoted idea by KZfaq pundits that when you attach a full-frame lens to a crop sensor camera, you get more "reach" (i.e., my 200mm telephoto becomes a 320mm telephoto on Canon APS-C), which I assumed meant more magnification, and hence more detail. My thinking has changed on this and I now think it's only the field of view that's affected. Would you care to comment or point me to a video that clarifies this situation? Thanks for your work on this and other videos of yours. Happy New Year!

struggled a bit with the definitions part of this (particularly how microcontrast differs from resolution when they're both expressed in the same units) however what I'm left with is that the glass is the glass. It doesn't change optical properties depending on what it is attached to. Perhaps a better metric for lens sharpness would be an estimate of the pixel size below which the gain in image detail is below a certain threshold. For example my Tamron 18-270 almost resolves adequately at a 6.7micron pp but would not resolve much more at 4.3 micon, so its critical sharpness threshold would be somewhere between.

I always drill down to measurements and then sharpness to see their results at each f stop and focal length. However their focus now is on cellphones. I doubt they will ever test the 90D with multiple lenses.

Great video, as usual, Michael. I like to look at DXO's charts for lens sharpness at given focal lengths and apertures, but the other "scores" introduce too much of the human estimate factor for my liking. Also, if DXO is not listing Nikon lens results with the D850, how many years are they planning to wait for that?!

MORE!!! Video is too short. Hopefully create a series.....?

Their "sharpness" value is a weighted average of how much resolution the lens can resolve when mounted on a specific body with a specific sensor, so it will never exceed the sensor resolution. I guess it makes sense to use the highest resolution sensors to run those tests.

Good to point this out, but I already interpreted it the way you say, it would be nice to have more details about the exact procedure dxomark uses to come up with this scores. Another case when the dxomark score does not seem to reflect real life performance is the Nikon d5 and in particular its high iso performance , dxomark does not give a great score but the photographer Steve Perry both through real life performance and through his tests thinks it is outstanding.....

Hi Michael, good work, but: The difference in sharpness of the same lens you mentioned on both d810 and d500 make sense in dxomark chart, as a physicist, we learned that the resolving power of a lens id dependent on the sine of an angle, this angle is the conic angle which light makes when going through the lens to project on the sensor, now: because the full frame sensor is bigger , the conic angle will be bigger, and mathematically , it's sine will be bigger, and hence the resolving power is directly proportional to the sine of that angle, the same lens will be more capable to resolve on the bigger full frame for the same focal length. At least,that is theoritical. Resolving power id dependent on wave length, inversely proportional, but it wont make any difference because shooting should be done in same light conditions.

Resolution of a camera in the film days was expressed as line-pairs per mm that could be clearly distinguished.Each pair consisted of a black and a white line. This is a linear (one dimension) unit. The reason behind this is, that the linear unit best relates to human perception of detail resolution (from detail, I distinguish gradation resolution, dynamic range and usable ISO, next to to this). In those days, good lenses and color films ranged generally between 80 and 120 line-pairs per mm. Say an average of 100. If we would equate one line in a pair with one "pixel" (mind you, a pixel is a virtual thing, not a physical thing), you would get a resolution of 7,200 x 4,800 (but it does not work exactly like that). The main message here is, detail resolution is a linear thing and megapixels are an area thing. If you want to compare the detail resolution between sensors, the simplest, bit naive, way is to compare the "pixels" on one side. Compare a 6,000 sensor with a 7,200 and in theory, the 7,200 has 20% more detail resolution. Mind that in megapixel area, we went from 24MP to 36 MP for a 3:2 sensor, which sounds like 50% more. Yes, it is 50% more in file size and potentially exponentially more processing to get raw file. But only 20% more detail resolution. What is the relevance of the megapixels, bottom line? Well, more is more. But, it primarily indicates how much processing needs to go on in a digital camera between reading each of the analog photosites (photovoltaic cells with each, either, a red, green, or blue filter on them) and AD-converting this into the "invented" RGB values assigned to each thing we call "pixel" and that we then think of as a raw file. The conversion of analog photosite readout to RGB pixel is done in an extrapolation that relies on interpolation or convolution or another smart algorithm to guess how much green and blue must be assumed with the red readout of the cell readout we have under scrutiny. The "Moiré" patterns that we saw a lot in the past were caused by the RGB filters being arranged in a fixed pattern and the algorithms not being smart enough. The algorithms got help physically, from creating unsharpness at low level that can be done away with if we increase detail resolution relative to the subject, not the FX to crop! My perception of P-MP is they factor all these things into that number. For pure detail resolution, they have "sharpness". DXO get to their numbers by buying the hardware they list and testing it. Not because they want to be nice to the public or critical of manufacturers, but because they make a camera raw processing application that you can buy. In the AD conversion of the readout of a photovoltaic cell (photosite) the question is if the manufacturer of a camera used a linear AD conversion or applied some non-linear function in the conversion, like an exponential/logarithmic or potentially some parabolic function. Imagine, we could assign the numeric range of each bit in the recorded bit-depth to a matching F-stop in the dynamic range we try to attain. Or, we could assign a decimal integer value to a lumen level we measure. This gives totally different results. Mind that the human eye/brain vision is assumed to work with 10-base log light values. Yes, "raw" is totally fake, to stick with populist US idiom. And DxO actually explain what they do, in their website. I have no relation with, nor interests in, any commercial entity in photography.

A crop is a crop hence the name crop factor not perceptual factor. Don't get how people get confused on this.

I must add that neither the distance, nor the lines and fine details in your test image were tight and tiny enough to prove that they were haveing the same sharpness, especially we rely on our eyes in such test. Thats what I think .. and that is what optics proved.

Aloha Michael, howzit... OK you are probably going to hear from Tony Northrup on this.

I remember the "Angry Photographer" complaining the DXO perceived Mega Pixel as a trustful tool to judge cameras and lenses. This happened already 3 to 4 years ago. I tried to understand the DXO mearsure, however I never got it. Finally I felt that this measure was invented by Marketing people who had never understood math and physics.

When I look at DXOmark scores, I'm always surprised at how low the they are. It seems that nearly all lenses and cameras are pretty much just crap. Only if you look at something really expensive like a 300mm f/2.8 L on a 5DsR do we see a decently high score. I wonder who’s paying DXOmark.

Interesting content Michael, thank you. Roger Cicala over at the extremely useful "Lensrentals" site has this to say about "Perceptual MegaPixels" (P-Mpix as DXOMark describes the idea): "...my major problem is it gives a number that takes into account: 1- lens performance at all apertures, 2- lens performance all across the field, 3- for a zoom, performance at all focal lengths, 4 - a formula of calculation that is not shared (but apparently is normalized down to a lowest common denominator, than multiplied up to a sensor in question, which is pretty insane), 5- a method of measurement that isn't shared and cannot be confirmed or denied by anyone else, and 6 - for lenses that work on multiple cameras and mounts the calculations are done on one camera and mount and therefore aren't valid. So we have this number that equates to something but we aren't sure what, based largely on measurements that we definitely don't know what, and people are making purchasing decisions based on that. More to the point, people are misinterpreting it to say things like 'this lens is rated at 16 megapixels so if I buy a 42 megapixel camera my images won't improve' when, in fact, they will."

Michael, not helpful to fan the flames of the "flat landers" on an issue laid to rest many moons ago. The truth is and obviously always has been that P-MPIX is not a lens sharpness metric but a perceived sharpness measure of a unique lens sensor combination which is clearly delineated by DXO. The greater the sensors resolution the greater the "potential" sharpness an image will be perceived to be when combined with a lens that is capable if resolving said sensor. Your demonstration of the obvious failed to point out that the D850s 46mp image would have been vastly sharper than the 24mp image of the D500 when using the same lens, exactly what DXO's metric would have predicted. Not rocket science but apparently quite far above biological science.

This is the only website who claims to make objective measurements, but does not disclose the methods they used to reach their "scientifc" conclusions.

I have had my doubts about DXO for a long time. I am not an academic. I am not a scientist, so couldn't prove anything. And although I like gear and tech talk, in the end, the only thing that counts for me, is if I am happy with the shots I took and the end results.

Learned friend, I would add to about being able to explain the data's source and processing, data may provide you with predictable measurements. But not a valid, nor relevant, theory, a priori.

Good info! Thanks.

I see what you’re saying, and 100% agree that a lens doesn’t change whether it’s on fx or dx, but we need to open the bigger can of worms and compare similar framing, because framing is a huge real world factor in the final resulting image. Nobody uses a 24mm on full frame and then says ‘I can get the same detail from my crop sensor’ and take the same image on dx but that’s cropped in 1.5x... because it’s not the same resulting frame (unless it’s a unique case like wildlife where you’d have to crop the ff image at least 1.5x anyway). And when you take comparable images at comparable focal lengths, apertures and ISOs with same megapixel sensors fx vs dx, full frame images tend to turn out sharper, so I think there’s merit in dxo saying that the dx sensors produce lower ‘detail’ with the same lens, because in the real world, comparible framing example like Tony N used in a recent video, the images are distinctly sharper on ff.

Tony N vs Michael the M... let the war begin

ha! finally someone to call their BS and bias!

dxomark ...full fake