I just wonder what the exactly influence of the sensor is on the quality of the lens. Difference due to pixel density can be calculated. But nobody can tell me that a whole serie of camera’s has to be fixed to do individual test.
It’s like with Nikon when they claim that exposure is set with the help of a database with 10000’s of pictures. I mus say I don’t hear that lately anymore.
George
Different sensors have different filters (low-pass, UV, IR, …), which affect sharpness. Crosstalk is yet another topic…
Google for “sensor microlenses”. Each sensor diode has a microlens on top of it to better concentrate the light, and it is particularly important on peripheral areas. Their design can change the rendition of a lens A LOT, especially concerning sharpness, chromatic aberrations and light falloff.
When teh camera is tested the results are stored in a kind of mathematical way so it can be used in pl. The same for the lens. I don’t see that when you know the characteristics of them why it’s impossible to combine them.
What exact aspect of a combination can’t be archived with this knowledge?
George
Absolutely non as new cameras tested can use all the exiting lenses tested for that make if csmera. But there is really why any lense camera combinstion couldn’t be used used with in logicsl limits. Full frame should be OK on 1/2 frame not the other way round. This would overcome the use of adapters taking lenes from many makes onto onother makes camera.
George (above in this thread) essentially is correct, although DxO will not discuss any details of the actual methods (measurement, mathematics, programming for implementation and end-user result). The DxO measurements are based upon what DXOMARK does. Details are a business secret. However, reading what DxOMark does, the requirement to test every feasible lens on every body that could accept such a lens is extremely time intensive; even much larger and better funded competitors cannot do that. Moreover, the use of robots as the technicians making the measurements is not yet feasible, as the training/programming of a robot to manipulate the controls of each lens and body still would take time. The most likely scenario is that DxO uses the mathematical method of convolution to compute lens A to body B. The actual data that ideally would be a tensor field is kept discrete with a finite number of elements (how many? trade secret) and then numerical integration is used to evaluate the convolution (again, trade secret). The results of the convolution are converted into the values of the lens plus body correction modules in PL, etc. The data structure and encodings in a correction module again are a trade secret. Presumably any AI is trained on many images not necessarily for the specific body plus lens. Even this procedure takes time and technicians (including programmers). The intellectual components of the process are in the mathematics and software design (including the choice of data structures as well as optimisation for execution time in the end-user application, such as PL8 Elite Complete). All of these details are trade secrets; I have not been able to find any research papers from DxO that provide significant detail of the actual practice. I do know from direct experimentation that PL8 Elite Complete outperforms open source GIMP for production workflow. For simple intake sorting, I no longer use Adobe rental Lightroom, but currently FastRawViewer as PL is too slow for this purpose.
The funny thing is I can’t find the Z6III on DxOMark at all. In PL it is known in combination with about or more then 100 lenses, several pages of 10.
George
There are several possible answers to your comment. First, examining DXOMARK, one finds that the public website reviews are several years behind current camera and lens releases. Thus, either DXOMARK is not able to stay “current” or DXOMARK is no longer making the results of their tests timely. Second, if DXOMARK is conducting tests on current production equipment, DXOMARK may be licensing those results to DxO (and perhaps others). Third, DxO may have established a testing facility independent of DXOMARK to keep current. The third alternative could be somewhat costly in initial infrastructure/equipment, particularly if DxO does not have access to the manifest of the infrastructure/equipment in the DXOMARK testing facility. This also would require trained technicians and perhaps engineers to oversee the facility. DxO is very unresponsive about these details (trade secret?). The issue for me is very simple: if DxO PL supports the bodies and lenses that I use to produce raw image files, PL Elite Complete current typically is very efficient workflow for post, particularly if I have a relevant PL preset of my own (I have a selection). The correction module/s produce images that have a high client acceptance fraction. Manual correction takes much longer and does not always produce as good a result.
DxoMark has become a joke. It used to be a decent source for lenses/camera reports (although I have always found Photographylife and Cameralabs much more useful and closer to my personal experience).
Nowadays, they are only interested in publishing reviews of smartphones. Probably, that’s what makes more money.
Although DxOMARK may not be current and provides public quantitative ratings that go well beyond simple MTF graphs for lenses, I suspect the techniques used by the DxOMARK testing facility are the source of the data used by DxO. DxO is not transparent as to whether DxOMARK data currently is used, if DxO contracts with DxOMARK for data on current equipment, or if DxO has a separate testing facility and (internal) data base for optics modules (camera plus lens plus TC, if used). I suspect that one or both of the latter are the reality, as DxO does show marketing material on DxO evaluating gear for the data used in DxO optics modules. As with Adobe (but not GIMP), all of the DxO program source code and the specifics of optic modules “construction” are “trade secrets”.
In responding to references about the involvement of DxOMark with PhotoLab lens profiles in 2019, 2021, and 2022, @Marie has continuously pointed out that DxOMark and DxO Labs have been two completely separate organizations since 2017. New lens profiles for PhotoLab are created by DxO Labs, not DxOMark.
Mark
Officially, two separate business entities without a single point of control (eg, not parts of a larger parent corporation). However, (1) it is not clear if DxO might not contract with DxOMark to produce the data used for the lens profiles in the DxO PL and related workflow applications. This data may not appear in either the camera or lens “ratings” of DxOMARK if produced solely for the use of DxO. (2) DxO might thus have to replicate the DxOMARK testing facility to produce the data used in the optics modules that PL requires for full functionality. (3) Some non-DxO/non-DxOMARK facility might be producing the data. The answers from Marie and the DxO marketing material does not make this clear. The methodology of taking data and then producing an optics module also is “secret”. Thus, I still suspect that neither DxOMark nor DxO actually had/has data indivdiually for eacnh of the lens/TC/body combinations for which there is a DxO optics module. Rather, I still strongly suspect that a convolution method is used to combine separate or limited combination data into an optics module for camera body A with lens B and perhaps a TC. If this is the case, then DxO could produce additional optics modules from existing data and release these to DxO workflow application users. I am not complaining, even given the secret methods DxO uses; but it is annoying not to have the optics module I might need for a specific body/lens combination that I use.
DxO existed before DxOMark. The first version of Optics Pro, the predecessor to PhotoLab was released in 2004. If I remember correctly DXOMark was not created by DxO until 2008 with the purpose of testing and ratng lenses. They separated as distinctly different companies in 2017. I wonder why people think that DXO Labs doesn’t have the ability to test their own lenses for the purpose of creating profiles.
In private conversations I have had with a few staff members over the years when DXO was much more active on this site, it was made clear to me that there has been no relationship between DXOMark and DXO Labs since 2017. You can continue to believe whatever you wish, I base my opinion on what I’ve been told.
Mark
Thank you for the information that is not public. Evidently, DxOMARK started from DxO developed measurement and testing techniques. From your reply, one may assume that DxO does indeed have a testing facility to develop optics modules for DxO workflow. Nonetheless, the number of combinations from each camera body, each lens available for that body, and any additions such as a teleconverter, makes a set of direct measurements for each separate lens/body problematic – not enough persons, not enough time. Thus, I still strongly suspect that the internal secret DxO methodology to develop an optics module for a specific body/lens/TC does use convolutions.
There have been recruitment ads for roles on testing and measurement by DxO, so I don’t see any reasons to not believe they are doing their own work.