Is it suppose to look like this... (Artifact banding on 48MP on Mini 3 Pro)

[frankjxl]

Hi all, I just received my first drone (Mini 3 Pro) to do some stress testing with the image quality, and I saw some issues with the 48MP files in Lightroom. In these examples, the water ripple seem to exhibit vertical lines/banding in decent lighting conditions.

I'm trying to understand if this is an issue with the debayer algorithm, or is it a problem with the sensor. It's hard to interpret since I haven't seen this exact issue come up from other sample images taken by other people.

Full size photo: original.jpg

I've also attached a few 100% zoomed version below so that you can see the full effects.

 

[Phantomrain.org]

Hi all, I just received my first drone (Mini 3 Pro) to do some stress testing with the image quality, and I saw some issues with the 48MP files in Lightroom. In these examples, the water ripple seem to exhibit vertical lines/banding in decent lighting conditions.

I'm trying to understand if this is an issue with the debayer algorithm, or is it a problem with the sensor. It's hard to interpret since I haven't seen this exact issue come up from other sample images taken by other people.

Here is the full size photo, I've also attached a few 100% zoomed version so that you can see the full effects.

View attachment 149921View attachment 149922

Hard to tell without seeing the whole picture first but this looks like low light conditions and or an area of shadows all of which can effect the picture .

Try posting something with a little more light or the full picture.

Phantomrain.org
Gear to fly in the Rain. Land on the Water.

 

[frankjxl]

Hard to tell without seeing the whole picture first but this looks like low light conditions and or an area of shadows all of which can effect the picture .

Try posting something with a little more light or the full picture.

Phantomrain.org
Gear to fly in the Rain. Land on the Water.

Full size photo is in the original post, link is on the last sentence. In the meantime, I'll try to replicate with more exposed photos.

 

[Phantomrain.org]

Full size photo is in the original post, link is on the last sentence. In the meantime, I'll try to replicate with more exposed photos.

This photo looks pretty good to me, Cleaned it up in lightroom and it Popped. Water looks good and colors enhanced and shadows came clean. Maybe add some sharpening and its a Nice photo.



Phantomrain.org
Gear to fly in the Rain. Land on the Water.

 

[brett8883]

Hi all, I just received my first drone (Mini 3 Pro) to do some stress testing with the image quality, and I saw some issues with the 48MP files in Lightroom. In these examples, the water ripple seem to exhibit vertical lines/banding in decent lighting conditions.

I'm trying to understand if this is an issue with the debayer algorithm, or is it a problem with the sensor. It's hard to interpret since I haven't seen this exact issue come up from other sample images taken by other people.

Full size photo: original.jpg

I've also attached a few 100% zoomed version below so that you can see the full effects.

View attachment 149921View attachment 149922

You’ve got to remember that with a high resolution image when you crop into 100%, you are cropping into a smaller tiny percentage of the actual sensor area because the percentage crop is determined by the difference between the image resolution and the resolution of your monitor which does not factor sensor size into the equation.

You are so far cropped in right here that of course you’re going to see noise and banding, this is normal. Your sensor size, which is what ultimately determines this, has not change size and is still comparatively small. Even cropping into 100% on a 48mp image from a full frame sensor you are bound to see some artifacts and the mini 3 sensor is a tiny fraction of that size. 48 megapixels is high resolution by any standard, full frame or even medium format. You cannot expect to see a perfect image when cropped in that far on that small of a sensor.

Cropping into 100% might have made sense when cameras were all in the 12 to 24 megapixel range because that crop represented roughly the same percentage of the sensor area. However, now with 48,60, and 100 megapixel sensors this no longer makes sense. At 48 megapixels you are looking at what was a 200% crop on a 12 megapixel image where, of course, you would expect to have noise and banding which is what you are seeing.

 

[Cymru]

Looks normal - the 48mp does have severe artefacts on bright, high contrast areas and patterns. Water speckle in particular.

Stick to the native 12mp and the problem goes away.

It does seem to be the debayer algorithm at fault.

Examples:-

Can see here its not just a case of things only noticeable because its 48mpixel - these cover large areas of the image.

Also, water reflection seems worst of all:



The 12mp images of the identical scene are clean and lack all these.

 

[Vic Moss]

Quad Bayer systems extrapolate data (basically), so you'll have more instances of jpg artifacting and other anomalies.

This is why I always tell people that these cameras are 12MP cameras that also happen to produce a 48MP image, not 48MP cameras.

 

[Dale D]

This photo looks pretty good to me, Cleaned it up in lightroom and it Popped. Water looks good and colors enhanced and shadows came clean. Maybe add some sharpening and its a Nice photo.

View attachment 149944

Phantomrain.org
Gear to fly in the Rain. Land on the Water.

The edit is good, but I would open the shadows even more!

 

[brett8883]

Quad Bayer systems extrapolate data (basically), so you'll have more instances of jpg artifacting and other anomalies.

This is why I always tell people that these cameras are 12MP cameras that also happen to produce a 48MP image, not 48MP cameras.

That’s not really true. It has 48 million pixels so it is a 48MP camera, that really isn’t debatable, it’s just the filter arrangement that is different from a traditional bayer filter.

All bayer filter sensors extrapolate data so even that isn’t unique to quad bayer. Quad bayer does stretch this a slightly further but we are now seeing larger sensors with quad bayer filters and this doesn’t have a noticeable affect on image quality.

Now on the other hand the 12 MP mode oversamples from 4 pixels which reduces noise so you do see an improvement in noise from the 12 MP mode at the pixel level.

 

[frankjxl]

I think I'm just mostly freaked out by the way that the artifacts are showing up. In most of other people's images, banding occurs as green and magenta fringing. But I haven't see another example where it has vertical lines going through the image. Could this be sensor variation that warrants a replacement?

 

[brett8883]

I think I'm just mostly freaked out by the way that the artifacts are showing up. In most of other people's images, banding occurs as green and magenta fringing. But I haven't see another example where it has vertical lines going through the image. Could this be sensor variation that warrants a replacement?

View attachment 150048View attachment 150049

It looks like some kind of computational sharpening or some other processing. It's expected from a camera like this. I just want to illustrate for you real quick how far you have zoom in to see this and then I'll shut up.


Your original "100% crop" is into 2.01% of the original image or less than 1 mega pixel. Your next crop in is 0.04% of the original image which is less than 0.2 mega pixels. Those little lines are at the pixel level, so we are now on a micron scale as it relates to the size of the sensor. Im just giving you trouble so don't take it any other way but come on man ain't nothing gonna look good at 0.04%.

It's not an issue with the sensor. If you want to call it an issue it's an issue with DJI's processing pipeline but the nature of small sensors like this is the lens cannot resolve 48 MP onto that small of an area so they use computational techniques to bring out more detail, quite successfully from the look of it.

The photo looks nice and no one will ever in a million years be able to see that. Export your photo at 20 MP or something if it really bothers you.

 

[LA 2020]

Hi all, I just received my first drone (Mini 3 Pro) to do some stress testing with the image quality, and I saw some issues with the 48MP files in Lightroom. In these examples, the water ripple seem to exhibit vertical lines/banding in decent lighting conditions.

I'm trying to understand if this is an issue with the debayer algorithm, or is it a problem with the sensor. It's hard to interpret since I haven't seen this exact issue come up from other sample images taken by other people.

Full size photo: original.jpg

I've also attached a few 100% zoomed version below so that you can see the full effects.

View attachment 149921View attachment 149922

Don't expect drone still pictures to be as good as a DSLR they aren't in that league yet (consumer drones). Get a program from Topaz called DeNoise AI it is almost magic when dealing with noise. I have saved many photos using this program and it's fast. Only around $60

 

[miemo]

I've also attached a few 100% zoomed version below so that you can see the full effects.

View attachment 149921View attachment 149922

I've seen exact same artefacts on the exact same situation: water ripples with light wind. On the other hand, haven't noticed it with any other subject. So seems to be limited to details of certain frequency & randomness.

I'm pretty sure this is due to the interpolation algorithm going from 12MP to 48MP. Guess in theory tweaking the algos with a firmware update could fix this, but then again might be really tricky to accomplish.

And yeah, has nothing to do with the chromatic aberration, i.e. the purple fringing, which is an easy fix e.g. in Lightroom. This upsampling artefact however would be pretty hard to fix in post, but luckily it shows up in such small details that in practice it would render itself invisible on most use cases.

 

[EpicFlight]

That’s not really true. It has 48 million pixels so it is a 48MP camera, that really isn’t debatable

It is very much debatable.

It has 48M distinct light sensitive sensors laid out in a rectangular grid. Those are not pixels.

Particularly with color imagery, those are not pixels. If they were, no Bayer filter would be necessary, and each location would capture red, green, and blue intensities at that location. That's not what happens.

 

[brett8883]

It is very much debatable.

It has 48M distinct light sensitive sensors laid out in a rectangular grid. Those are not pixels.

Particularly with color imagery, those are not pixels. If they were, no Bayer filter would be necessary, and each location would capture red, green, and blue intensities at that location. That's not what happens.

Well whatever you think they should be called, there are 48 million of them just like you’d see on a traditional bayer sensor that produces 48MP images. I don’t think we have a disagreement about that.

 

[EpicFlight]

Well whatever you think they should be called, there are 48 million of them just like you’d see on a traditional bayer sensor that produces 48MP images. I don’t think we have a disagreement about that.

A traditional sensor has 4 sites per pixel.

The sensor, filter, and some math approximate the image data (RGB) for each of those sites to produce a pixel (picture element), but it's calculated, and error prone. It produces an image with 48M pixels, not a 48M pixel image. To do that you would need a sensor with 256M photosensors and a traditional R2GB Bayer filter.

 

[Flashheart]

Those artefacts seem to only appear on reflective surfaces. Reflective surfaces scatter light. There is a reason why those artefacts have a magenta tone.
Regardless, those artefacts should disappear if you put a CPL filter in front of the sensor.

 

[brett8883]

A traditional sensor has 4 sites per pixel.

The sensor, filter, and some math approximate the image data (RGB) for each of those sites to produce a pixel (picture element), but it's calculated, and error prone. It produces an image with 48M pixels, not a 48M pixel image. To do that you would need a sensor with 256M photosensors and a traditional R2GB Bayer filter.

Please define “traditional sensor” and please give an example of a modern consumer digital camera that uses this approach. And please, by your definition how many pixels does the sensor on the Mini 3 Pro have?

 

[test2000Anafi]

it does not need to be better than DSLR, which it is in a lot of ways; you cannot throw your DSLR in the sky with a timer and hope it takes a good picture, it merely needs to be better than anything below Mavic 3.

You can even pixel peep mavic 3 and be disappointed btw, nothing to do with Mini 3 Pro really

 

[EpicFlight]

Please define “traditional sensor” and please give an example of a modern consumer digital camera that uses this approach. And please, by your definition how many pixels does the sensor on the Mini 3 Pro have?

A traditional bayer sensor. Not a Quad Bayer.

Maybe you didn't know there have always been 4 photosites for every pixel, with an RGB filter over each set of 4 compact sensors. Traditionally this filter is a square divided into 4 subsquares two adjacent green, with the other two red and blue. That's a Bayer filter.

That's how it's always worked. Camera sensors with 10MP have 40M photosensitive sites.

Some clever engineer realized that with TWO separate sites collecting green intensity near each other the difference can be used, with the red and green intensities, to estimate, with error, the RGB values for each of the 4 photosites. By arranging the 4 filters with the two green diagonal instead of adjacent, the error is reduced.

Hence the introduction of the quad-bayer filter, and claimed higher resolution images. The semiconductor industry didn't have a 4x leap in density all of a sudden.

If you want the highest quality, sharp, color correct images, do not shoot in a quad bayer mode.