What is the optimal pixel count on a cropped sensor for Canon 100mm macro USM lens?

While choosing used camera to buy, I was checking some combinations of camera bodies with Canon 100mm macro USM lens at DXOMARK and noticed that sharpness (in perceived megapixels) looks very strange. It reads:

| Camera | Sensor Mpix | Perceived Mpix |
| ------ | ----------- | -------------- |
| 500D   | 15          | 8              |
| 7D     | 18.84       | 10             |
| 550D   | 18          | 11             |
| 100D   | 18.1        | 11             |
| 70D    | 20.17       | 12             |
| 700D   | 18.1        | 12             |
| 750D   | 24.23       | 11             |
| 760D   | 24.23       | 11             |


What is going on? Does the lens perform worse on 750D than on 700D, for example? Or how these figures should be interpreted? Should we assume that sharpness will become even worse with some sensor with higher pixel count?

• Those variations are all within the probable margin of error of the testing regimen. In other words, they mean absolutely nothing. Nov 25, 2022 at 19:18
• That is to say, the variation from one copy of a 750D to another copy of a 750D could be as great or greater than the variation between one copy of a 750D and one copy of a 700D. Nov 25, 2022 at 23:34

DxOMark's "perceptual megapixel" rating is an attempt to distill a lens's MTF charts, which describe the lens's optical resolution of sagittal and tangential rays as a function of distance from the lens's optical center, into a single number or rating, when that lens is mounted on a particular camera body.

Distilling something as complex as a complex or 2-valued continuous function into a single number is problematic, and discards a lot of information that is useful for the discerning buyer.

Answering your question, there is no "optimal pixel count" of a camera body for a particular lens.

• While there may not be such thing as an "optimal pixel count", there is a point at which having higher resolution (pixel size or count) is irrelevant because of the limits of the lens resolution - e.g.. w/ totally made up numbers - if the lens can only provide N lines/mm resolution, having a sensor that provides 2N lines/mm doesn't really help much outside of antialiasing considerations.... Nov 23, 2022 at 19:34
• @twalberg You're not wrong at all. However, thinking about the situation in the larger picture: nobody limits the new camera they are buying based on the performance of the subpar, probably older, lens, or are thinking of designing their system around. In film days, the camera didn't matter regarding resolution; it was the film, the replaceable sensor and recording element in one, that mattered there. While ergonomics, lens mount, etc., were definitely all concerns, ultimately the camera body was just a dark box. But now, in DSLR days, the sensor & software are of significant concern.
– scottbb
Nov 23, 2022 at 20:46
• @scottbb That must be totally right in most cases, I guess; but I mainly use my camera with two or three lenses, and don't plan to change them anywhere soon. So the main question for the camera upgrade is (besides the pure ergonomics - like a tilting/articulated display) is whether my primary lens can benefit from its resolution. So I'm trying to understand if the said 100mm macro would perform any better with 18/20/24 Mp cropped sensor, and what figure is the last one giving some additional details so to say. Nov 24, 2022 at 15:49
• @NickL How you use such a lens could also affect which specific body would perform better with a specific lens. Are you shooting Macro at MFD? Or using a lens optimized for flat field reproduction at MFD to shoot more distant 3D scenes? Nov 25, 2022 at 23:26

Distilling a large amount of data down to a single number is inherently a compromise in terms of comparing two different lenses.

For example: One lens may be sharper at the center of the field but not as sharp at the edges and corners. The other lens may be slightly softer at the center of the field, but maintain that sharpness much better towards the edges and corners.

Even with the same lens, as is the case in your question, differences in camera design can affect results.

Some of the things, other than sensor resolution, that can dramatically affect the acutance of images taken with various cameras:

• The presence (or not) and strength of the low pass filter in front of the sensor. These are sometimes called anti-aliasing or moiré reduction filters. The Canon 7D, for instance, was known to have a strong low pass filter compared to the 50D that preceded it and the 7D Mark II that followed it. Images from the 7D, even those made in controlled laboratory conditions, suffered from the stronger filter. On the other hand, video footage from the 7D (which is inherently lower resolution that 20MP still images by a factor of 10) looked better with less moiré. The 1D X Mark III has a different type of newly developed Low pass filter. Compared to the 1D X Mark II, which had the same basic sensor with a different filter stack in front of it, the Mark III performed noticeably better. The 45MP EOS R5 with the same kind of newer technology low pass filter beats the older 50MP EOS 5Ds with the old style low pass filter.
• The thickness(es) and refractive index(es) of the cover glass(es) in front of the filter. The cover glass may include coatings for infrared and/or UV cut filters, or these filters may be separate ones in the stack in front of the sensor.
• The design and manufacturing quality of the microlenses in front of the Bayer mask.
• The properties of the Bayer mask itself.
• Processing and amplification of the analog signal from the sensor, the bit depth of analog to digital conversion, and the processing routines applied to that digital information.
• Even two copies of the exact same camera model can test with slightly different results due to the variations caused by manufacturing tolerances, especially with regard to the distance and squareness of the sensor with the camera's lens mount ring.

If you want to get the most acutance out of any lens that projects an image circle large enough for a full frame camera, then use a full frame sensor. The reason for doing so is the difference in enlargement ratios between a full frame sensor and an APS-C sensor when viewing images from each at the same display size.

These differences between enlargement ratios due to sensor size are significantly greater than the minor differences between different cameras with the same sensor size.

Why is FF sharper than crop body for the same framing of the same object?

When you enlarge the image by 1.6X more linearly, which is 2.56X more by area, you also enlarge all of the blur in the image from the smaller sensor that much more by the same factor. Some of the blur that can't be perceived as blurry by the viewer when enlarged at the lower enlargement ratio required to view a FF image at a certain size will be perceived as blur at the higher enlargement ratio required to view an APS-C image at the same display size.

To get the same acutance from an image enlarged by 1.6X more, you need a lens that can resolve 1.6X more lines per millimeter.