50 MP photos not real?

[Landey]

Hi there!

I just had a closer look at two photos I took a few days ago - one taken with 12 MP, the other with 50 MP picture resolution.
I resized the 12 MP photo to the exact size of the 50 MP photo - using Affinity Photo, a serious photo editor.

Just see the results: Do you see any significant differences between the image sections I picked?
For me, it looks like the option for taking 50 MP photos does NOT take 50 MP photos at all - just grabs a 12 MP picture and saves it as a 50 MP picture, just like you can do with any post-processing software. There's just tiny differences (slightly lower histogram and contrast values), likely to be caused by minor changes of lighting conditions as the photos were taken about 20 seconds apart.

What's your opinion?
Please also do your own tests and share the results. Let's find out if it's the same everywhere or just a "specialty" of my Nano.
Autel: Your comments would be highly appreciated.

 

[Brock Inspector]

Resolution is an objective thing, but it's subject to a lot of factors. One factor is whether light rays can be bent around within lens elements and come out the other side intact. No one has so far proven (at least not to my satisfaction, a professional photographer of 30+ years) that a drone's or smartphone's small sensors have lenses and optical systems that can resolve more than an effective 12MP in a larger sensor, no matter how high the pixel count. Light has to bend a lot at these extremely short focal lengths, and diffraction hits hard. There are other limiting factors, like random light wave scattering that happens in low contrast and low level light. (It's the reason almost all objective testing is done with USAF charts in 1000:1 contrast lighting. Speaking of testing, have you performed any actual tests?)

 

[herein2021]

Hi there!

I just had a closer look at two photos I took a few days ago - one taken with 12 MP, the other with 50 MP picture resolution.
I resized the 12 MP photo to the exact size of the 50 MP photo - using Affinity Photo, a serious photo editor.

Just see the results: Do you see any significant differences between the image sections I picked?
For me, it looks like the option for taking 50 MP photos does NOT take 50 MP photos at all - just grabs a 12 MP picture and saves it as a 50 MP picture, just like you can do with any post-processing software. There's just tiny differences (slightly lower histogram and contrast values), likely to be caused by minor changes of lighting conditions as the photos were taken about 20 seconds apart.

What's your opinion?
Please also do your own tests and share the results. Let's find out if it's the same everywhere or just a "specialty" of my Nano.
Autel: Your comments would be highly appreciated.

Resizing a 50MP image to the same dimensions as a 12MP image will always look identical when shot with the same camera because the very act of resizing the image reduces the effective pixel count to match the new resolution. The differences though should be readily apparent when you actually print out the images. Here is a good article on print sizes based on MP. This latest MP war is almost comical; more people are viewing content on mobile devices than ever before; 3MP is all you need for optimal cell phone resolutions yet people buy into the MP marketing. Video is the same way, most platforms restrict viewing video on mobile devices to 360P, yet camera makers are in a race to provide 6K and 8K since that's what the marketing department has sold to the users.

If you do want to retest your 12MP vs 50MP image digitally without printing it out a better test would be to resize the 12MP image to the same dimensions as the 50MP image then look at them side by side. The 12MP image should be much more pixelated than the 50MP image.

With that being said, I do think any 50MP image out of such a small sensor will fall apart quickly when printed or compared to a 50MP image shot with a full frame camera due to limited dynamic range and the number of photosites that they have managed to place on such a tiny sensor.

 

[izometric]

If you do want to retest your 12MP vs 50MP image digitally without printing it out a better test would be to resize the 12MP image to the same dimensions as the 50MP image then look at them side by side. The 12MP image should be much more pixelated than the 50MP image.

This is exactly what he said he did.

 

[Landey]

Thank you for answering, guys!

Just to put some things right:
I am well aware about optics, I do photography and astrophotography for quite some years, so I know a bit about the matter.
I also don't expect highest quality from devices with little sensors and tiny lenses - if I wish to take a quick and nice picture, I'll use my Huawei P40 Pro, but if I wish to take a GOOD photo, I'll use my Fuji with it's large sensor, excellent lenses and wide range of manual settings. And of course the Fuji with it's less-MP sensor always beats the more-MP sensor of the Huawei if you do things right.

The issue is what izometric already pointed out: If I take a 12 MP (12.5 to be exact) picture and scale it to 50 MP, it looks EXACTLY like a "genuine" 50 MP photo. That's what the pictures in my first post show.
I know that Autel does binning, thus taking four pixels of the 50 MP sensor and combining them into one for achieving better picture quality and gathering more light.
So there must be a noticeable difference between binned 12.5 MP pictures and native 50 MP pictures. But there isn't.
That's why I suspect the 50 MP setting doesn't provide 50 MP pictures, just a scaled up version of the binned 12.5 MP picture.

 

[herein2021]

Thank you for answering, guys!

Just to put some things right:
I am well aware about optics, I do photography and astrophotography for quite some years, so I know a bit about the matter.
I also don't expect highest quality from devices with little sensors and tiny lenses - if I wish to take a quick and nice picture, I'll use my Huawei P40 Pro, but if I wish to take a GOOD photo, I'll use my Fuji with it's large sensor, excellent lenses and wide range of manual settings. And of course the Fuji with it's less-MP sensor always beats the more-MP sensor of the Huawei if you do things right.

The issue is what izometric already pointed out: If I take a 12 MP (12.5 to be exact) picture and scale it to 50 MP, it looks EXACTLY like a "genuine" 50 MP photo. That's what the pictures in my first post show.
I know that Autel does binning, thus taking four pixels of the 50 MP sensor and combining them into one for achieving better picture quality and gathering more light.
So there must be a noticeable difference between binned 12.5 MP pictures and native 50 MP pictures. But there isn't.
That's why I suspect the 50 MP setting doesn't provide 50 MP pictures, just a scaled up version of the binned 12.5 MP picture.

My apologies, somehow I read it forward and interpretated it backwards long days and late nights tend to do that to me.

Have you tested at night to see if the binned version is better in lowlight? Also, was your output from the camera RAW or JPG?

It is definitely entirely possible that their firmware has a bug and is not producing actual 50MP images, are the file sizes significantly different? I would try taking a picture of a highly detailed scene like a test chart or even just a well lit page in a book in a room with it on the ground and zoom in to 400% to really compare. If there is still no difference then I would submit the results to Autel in a trouble ticket.

 

[Landey]

My apologies, somehow I read it forward and interpretated it backwards long days and late nights tend to do that to me.

Have you tested at night to see if the binned version is better in lowlight? Also, was your output from the camera RAW or JPG?

No problem at all - given the volume I tend to write, it's no wonder you overlooked something.

I did no night tests with that "50 MP" resolution as the daylight test already was a bit frustrating. I took some regular night photos, they are okay so far, but overexposed. Will try lowering the EV if that's possible with that kind of long exposure shots.
Output was JPG for both resolutions because the app doesn't even offer RAW output in 50 MP mode. Would love to see the RAW sensor output - but no way.

Files sizes: 12 MP: 9.4 MB, 50 MP: 22.4 MP, 12 MP picture scaled up to 50 MP using Affinity Photo: 22 MB.
So the file sizes of that original "50 MP" picture and the one I manually scaled up to 50 MP are almost identical.

I deliberately took a picture showing a lot of greenery with slightly different hues as I know this is a great challenge for digital cameras. There are also plain and high contrast parts in the whole picture - they also look identical, just the same.

 

[izometric]

I think they count the RYYB pixel array as 4 pixels. That's how they came to the 50 MPix number.

 

[Landey]

I think they count the RYYB pixel array as 4 pixels. That's how they came to the 50 MPix number.

I guess they are using an RYYB matrix with a quad-Bayer pattern, meaning that a square of four subpixels of the same color form one larger subpixel, four of them (one red, one blue, two yellow) building a full "main" pixel. - As far as I hopefully understood the principle.
So the native resolution is indeed 50 MP, but the color resolution is just 12.5 MP.
Autel COULD use the native 50 MP resolution to apply even more sharpness and details to the pictures, but this needs a special and quite complex algorithm.
I guess they don't use such an algorithm yet (at least not for saving 50 MP pictures), thus we obviously got 12.5 MP pictures upscaled to 50 MP for now.
So there might be some more coming.
IF Autel does it and does it right, they could even beat the slightly better detail resolution the Mavic 3 provides at present.

 

[izometric]

The Nano+ sensor is RYYB -> one red, two yellow, one blue.

RRYB (Red, red, yellow, blue?) -> do not exist.

About different sensor types: Color filter array - Wikipedia

 

[herein2021]

I guess they are using an RRYB matrix with a quad-Bayer pattern, meaning that a square of four subpixels of the same color form one larger subpixel, four of them (one red, one blue, two green) building a full "main" pixel. - As far as I hopefully understood the principle.
So the native resolution is indeed 50 MP, but the color resolution is just 12.5 MP.
Autel COULD use the native 50 MP resolution to apply even more sharpness and details to the pictures, but this needs a special and quite complex algorithm.
I guess they don't use such an algorithm yet (at least not for saving 50 MP pictures), thus we obviously got 12.5 MP pictures upscaled to 50 MP for now.
So there might be some more coming.
IF Autel does it and does it right, they could even beat the slightly better detail resolution the Mavic 3 provides at present.

I am not sure if it is possible to get the full resolution from the sensor with an RYYB array. Since 50% of the pixels can only receive shades of yellow, only 25% can receive red and the remaining 25% can only receive blue, I don't see how you would ever be able to create an image with 100% of the sensor resolution. This article performs a pretty good analysis on RYYB sensors and in the article they also mention it is not possible to get the full resolution from certain array types.

I think your best bet is to really test the lowlight capabilities more so than high resolution. The sole purpose of the RYYB configuration was to improve lowlight color sensitivity and from what I have seen online it does this very well.

It would be nice if the EVO III ships with the RYYB sensor in a m4/3 configuration with an effective resolution of around 25MP. Normally I do not care about resolution but with drone work specifically, my clients tend to use them in 3D renderings and building sized advertisements so the one place where I really need high resolution happens to be difficult to economically achieve.

 

[Landey]

The Nano+ sensor is RYYB -> one red, two yellow, one blue.

RRYB (Red, red, yellow, blue?) -> do not exist.

About different sensor types: Color filter array - Wikipedia

You are perfectly right, izometric!
I did a typo, RRYB instead of RYYB - also I wrote about green instead of yellow, which I will correct immediately.
Feedback likes yours is very important for correcting mistakes, so other pilots reading the post later will get correct information.
Thank you very much!

 

[Landey]

I am not sure if it is possible to get the full resolution from the sensor with an RYYB array. Since 50% of the pixels can only receive shades of yellow, only 25% can receive red and the remaining 25% can only receive blue, I don't see how you would ever be able to create an image with 100% of the sensor resolution. This article performs a pretty good analysis on RYYB sensors and in the article they also mention it is not possible to get the full resolution from certain array types.

I think your best bet is to really test the lowlight capabilities more so than high resolution. The sole purpose of the RYYB configuration was to improve lowlight color sensitivity and from what I have seen online it does this very well.

It would be nice if the EVO III ships with the RYYB sensor in a m4/3 configuration with an effective resolution of around 25MP. Normally I do not care about resolution but with drone work specifically, my clients tend to use them in 3D renderings and building sized advertisements so the one place where I really need high resolution happens to be difficult to economically achieve.

Hello there, herein2021!

Thank you very much for linking the GSMArena article. I knew it already, but it's a great starter for everyone interested in that technology.

I had the chance of experimenting a bit with a 50/12.5 MP sensor a while ago. The results very quite mixed: With some subjects I did not see quality improvements, with other subjects there were noticeable quality benefits. There was color fringing with some shots (like outlines of distant power poles against the darker sky), others showed a far better resolution. I also noticed distinct improvements with e. g. white or grey writings on plain-colored backgrounds.
If we take the native pixels, we can't say of which exact color some pixels should be, but we can clearly determine if there's light or not - if sufficient lighting is given. As the sensitivity decreases if the pixels don't get binned/combined, we need a bit more light for achieving a beneficial effect. Thus we get more resolution, but with the drawback of losing color information. It's also possible that we get no information at all - if the light reaching the sensor area cannot pass the sensor's color filter. This explains the better experience with white objects as white light passes all color filters. There are specialized algorithms trying to "guess" the correct color of a single, unbinned pixel, thus it might depend on the use of such algorithms if 50 MP indeed should lead to a better overall resolution.
But still, quality gain will strongly depend on the subject. Thus I decided to stick with 12.5 MP because the results of the 50 MP pictures just weren't reliable.

I just did a primitive test in a low-light situation, right in my bedroom.
I took the Nano and shot a picture of the very dimly lit scene, had a look at the metadata (ISO 400, 1/11 s) and set the same settings on my Fuji, using the typical but capable XF 18 - 55 lens, aperture wide open. The results were comparable, with the Fuji gathering a bit more light, the Nano providing more lifelike colors, both providing good sharpness, thanks to the OIS doing a great job.
Sure, this is no real test - I'll try some comparisons outside in real flight.

I know the low-light capabilities of Leica/Huawei cameras from the P40 Pro I use for years, they do very, very well.

Regarding resolution, I fully agree. Sure, resolution is not everything - but if you would combine world-class lenses with a 100 x 100 pixels sensor, I doubt the regular photo quality would exceed a 25 MP sensor with a lesser-grade lens.


To all other pilots: Sorry for that techtalk - but maybe there's some of you finding it still interesting.

 

[Darkmatterx76]

The question is, why aren't they allowing this in RAW?

 

[herein2021]

Hello there, herein2021!

Thank you very much for linking the GSMArena article. I knew it already, but it's a great starter for everyone interested in that technology.

I had the chance of experimenting a bit with a 50/12.5 MP sensor a while ago. The results very quite mixed: With some subjects I did not see quality improvements, with other subjects there were noticeable quality benefits. There was color fringing with some shots (like outlines of distant power poles against the darker sky), others showed a far better resolution. I also noticed distinct improvements with e. g. white or grey writings on plain-colored backgrounds.
If we take the native pixels, we can't say of which exact color some pixels should be, but we can clearly determine if there's light or not - if sufficient lighting is given. As the sensitivity decreases if the pixels don't get binned/combined, we need a bit more light for achieving a beneficial effect. Thus we get more resolution, but with the drawback of losing color information. It's also possible that we get no information at all - if the light reaching the sensor area cannot pass the sensor's color filter. This explains the better experience with white objects as white light passes all color filters. There are specialized algorithms trying to "guess" the correct color of a single, unbinned pixel, thus it might depend on the use of such algorithms if 50 MP indeed should lead to a better overall resolution.
But still, quality gain will strongly depend on the subject. Thus I decided to stick with 12.5 MP because the results of the 50 MP pictures just weren't reliable.

I just did a primitive test in a low-light situation, right in my bedroom.
I took the Nano and shot a picture of the very dimly lit scene, had a look at the metadata (ISO 400, 1/11 s) and set the same settings on my Fuji, using the typical but capable XF 18 - 55 lens, aperture wide open. The results were comparable, with the Fuji gathering a bit more light, the Nano providing more lifelike colors, both providing good sharpness, thanks to the OIS doing a great job.
Sure, this is no real test - I'll try some comparisons outside in real flight.

I know the low-light capabilities of Leica/Huawei cameras from the P40 Pro I use for years, they do very, very well.

Regarding resolution, I fully agree. Sure, resolution is not everything - but if you would combine world-class lenses with a 100 x 100 pixels sensor, I doubt the regular photo quality would exceed a 25 MP sensor with a lesser-grade lens.


To all other pilots: Sorry for that techtalk - but maybe there's some of you finding it still interesting.

That's why I asked about RAW earlier, if my suspicions are correct, they are only using the super pixels from the sensor to produce the RAW image and simply upscaling JPGs to get 50MP. This is definitely what I would call deceptive marketing at its finest. Examples of this are endless in the camera industry such as calling a sensor 1" or MFT when in reality neither are close to being accurate actual sensor sizes. Lets not forget the "high res" modes of MILCs using IBIS sensor shift technology which produces nearly indiscernibly better images vs the standard res versions or Autel stating the EVO II 6K can fly for 40min when it can't even leave the ground with more than 32min showing on the countdown timer.

I do think if you compare the Nano in a real world low light test against nearly any previous prosumer grade drone that you will find this is where it excels. My optimal low light settings for the EVO II 6K are F2.8, 1/30s, 3600K WB, ISO800, and LOG color profile. I wouldn't be surprised if ISO400 or even ISO200 would produce the same or better results in the same scenario.

High resolution was never going to be the strong point of such a small sensor but to state it can produce 50MP images is as deceptively false as it gets if indeed it is simply upscaling from 12.5MP. That's like saying a 20MP RGB sensor can take 60MP images even though you can never really use the R/G/B photosites individually for image production.

The question is, why aren't they allowing this in RAW?

Probably because the demosaicing algorithm can't properly interpret the individual photosites in the RYYB array in order to form a coherent image that can have the proper RGB values applied to it.

 

[Darkmatterx76]

OK, so, if we counted the YY as just 1 pixel, so 3 pixels total, how many MP's would the camera be? 12 seems too low a number.

 

[izometric]

Why would you count YY as one pixel?

RYYB are 4 physical subpixels which form one pixel that can detect all visible spectrum.

So it is clear that this is a marketing move:

Count all subpixels -> you end up with 50 Mpix.
Count all pixels -> you end up with 12.5 Mpix. (this is your usable resolution -> anything else is AI enhancing).

 

[Landey]

That's why I asked about RAW earlier, if my suspicions are correct, they are only using the super pixels from the sensor to produce the RAW image and simply upscaling JPGs to get 50MP. This is definitely what I would call deceptive marketing at its finest. Examples of this are endless in the camera industry such as calling a sensor 1" or MFT when in reality neither are close to being accurate actual sensor sizes. Lets not forget the "high res" modes of MILCs using IBIS sensor shift technology which produces nearly indiscernibly better images vs the standard res versions or Autel stating the EVO II 6K can fly for 40min when it can't even leave the ground with more than 32min showing on the countdown timer.

I do think if you compare the Nano in a real world low light test against nearly any previous prosumer grade drone that you will find this is where it excels. My optimal low light settings for the EVO II 6K are F2.8, 1/30s, 3600K WB, ISO800, and LOG color profile. I wouldn't be surprised if ISO400 or even ISO200 would produce the same or better results in the same scenario.

High resolution was never going to be the strong point of such a small sensor but to state it can produce 50MP images is as deceptively false as it gets if indeed it is simply upscaling from 12.5MP. That's like saying a 20MP RGB sensor can take 60MP images even though you can never really use the R/G/B photosites individually for image production.



Probably because the demosaicing algorithm can't properly interpret the individual photosites in the RYYB array in order to form a coherent image that can have the proper RGB values applied to it.

Absolutely second that!

What we presently get as a "50 MP" image is just a scaled up version of a regular 12.5 MP picture, with a color resolution AND pixel resolution of 12.5 MP.
A "real" 50 MP picture should provide 50 MP pixel resolution and a 50 MP color resolution interpolated from the 12.5 MP color information.
Easier put: A picture with 50 MP true brightness information and 50 MP "guessed" color information for every pixel.
The correctness of that color information depends on many factors, the subject being one of them. In an ideal situation, we could get 50 MP pixel and color resolution, which of course is up to the algorithms used for guessing the missing colors.
Thus sometimes we could call it deceptive marketing, sometimes not.
Similar with resolution enhancements gained with tiny sensor displacement using IBIS. In fact, the resulting resolution exceeds the physical sensor resolution. But it should be called a different way, not sensor resolution, just because it's not the resolution achieved at the same point of time. On the other hand: If that resolution enhancement is fully done within the time frame it takes for capturing a single shot ... Question of definition. In that light, I don't even dare to think of "rolling shutter".

There is deceptive marketing at countless places. Just think of that digital zoom pest, battery capacities, of course flight times, chargers charging three batteries at once, and the like. One manufacturer starts it, all others follow for not looking worse. And humans like to be deceived, also happily join the game. Just look at that five-stars reviews at Amazon for subzero products. Half of them is paid, but the other half consists of real buyers telling lies for patting themselves on the back.
In a world of liars, the ones telling the truth would be regarded as outlaws.

Compared with the DJI Mini 2, the Nano Plus really shines. Stills taken with good light sometimes look almost identical, but the video output really is a different class, in good and in adverse (low-light) conditions. At least that's my subjective feeling as I didn't fly the drones at the same time at the same place, yet.
Maybe I'll try today and share the findings with you.
But I'm not sure if I should go with Normal, HDR or Log mode with the Nano, just cannot decide. So every input regarding this would be highly appreciated.

Amen.

 

[Landey]

Why would you count YY as one pixel?

RYYB are 4 physical subpixels which form one pixel that can detect all visible spectrum.

So it is clear that this is a marketing move:

Count all subpixels -> you end up with 50 Mpix.
Count all pixels -> you end up with 12.5 Mpix. (this is your usable resolution -> anything else is AI enhancing).

Yes and no. Because it's the same with every sensor using a Bayer pattern - there's always 4 subpixels building a superpixel, thus the physical color resolution is always just 1/4, while the pixel resolution (brightness information) represents the native sensor resolution. And due to capable algorithms the results are pretty good, far beyond the theoretical 1/4.
Thus the detail resolution indeed is 50 MP, but the color resolution might be less.
Real hard deciding whether to call that a marketing move or not. Easy if shooting scenes only consisting of blue or red objects, but difficult with everyday scenes.

 

[izometric]

Then how do you call the 1 inch 20 MPix sensor from Air2S (which by the way, wipes the floor with Nano+)?

5 Mpix or 80 Mpix?