There still seems to be the view among some Mini 3 Pro users that the 12MP mode is superior to the 48MP mode for stills. But I just don't see it. I have tested it about 5 times now and everytime the 48MP appears more detailed. There has been some talk that you can usample the 12MP to be as good or better than the 48MP but you can probably do that to the 48MP image also.  At 100 iso there is some evidence that the 48MP has more noise but noise can be reduced and to my eyes is not an issue.
So I did another test. Here is a screengrab of the 48MP overall view.... I took an image of this in DNG mode 12MP and 48MP then using Photoshop converted straight to TIFF without any additional adjustments...

   


Here is a side by side comparison, left 48MP and right 12MP. Left is at 300% and Right is at 580% to equal the 48MP view...

compare

If anybody can show how you can make a 12MP look as detailed as a 48MP (with all its Bayer trickery) I would love to know as I am only after the best image quality I can achieve from this realtively small sensor.

On that if anybody is looking for an alternative DNG converter DXO is iminently releasing it's Mini 3 Pro Profile so will be very interesting to see how the TIFF files look from DNG once that has been released. You'll also need DXO Pure RAW 2 or DXO Photolab to use it.

Cheers

Bussty

Hello there Bussty. Good day and thank you for sharing these interesting information and photos with us. Again, thank you for your support and have a nice day.

DJI Stephen Posted at 9-1 21:24
Hello there Bussty. Good day and thank you for sharing these interesting information and photos with us. Again, thank you for your support and have a nice day.

Hi DJI Stephen always happy to explore how to get the best out of this drone

Bussty Posted at 9-1 21:27
Hi DJI Stephen always happy to explore how to get the best out of this drone

I'm not really convinced about the 48 MP images, Andrew. It may be that I got a camera that is not so insanely superior. But while I don't find much wrong with the 12 MP images, the result with the 48 MP after no further processing in ACR is not convincing. This shows a 100% crop of the lower right corner of a photo.



It may be that DxO can get a handle on this. But there would have to be some work done to eliminate the distortions at the edges and the color and sharpness problems.

Cheers, Bernd

SPLflyer Posted at 9-1 22:03
I'm not really convinced about the 48 MP images, Andrew. It may be that I got a camera that is not so insanely superior. But while I don't find much wrong with the 12 MP images, the result with the 48 MP after no further processing in ACR is not convincing. This shows a 100% crop of the lower right corner of a photo.

[view_image]

Hey Bernd

Do you have the 12MP version of that same corner and magnification to compare? I will admit the very corners aren't that great but thought that would have applied to both 12MP and 48MP as more about the lens?

I also think there is variability amongst lenses with some on this forum going back for a replacement.

Cheers

Bussty
P.S. What magnification is that as arent even beginning to see noise on that 300% 48MP I posted above?

Bussty Posted at 9-1 22:33
Hey Bernd

Do you have the 12MP version of that same corner and magnification to compare? I will admit the very corners aren't that great but thought that would have applied to both 12MP and 48MP as more about the lens?

Hi Andrew, I did not compare the same motif. Comparable 12 MP photos also show slightly increasing blur towards the edge, but at least they don't have any clearly visible distortions. Since I have so far almost exclusively built panos from them, I have not yet particularly noticed weaknesses in the 12 MP ;-).

As I wrote, the image above shows exactly 100% original size! No magnification!

Cheers, Bernd

Edit P.S. I did this in automatic mode: Exposure time 1/1500, ISO 100

Good comparison,   I did some shots which included people and the 48mp images certainly showed more details of the people than then 12mp,   Now all I need is an AEB mode for it,  I dont care if its slow.  

SPLflyer Posted at 9-1 22:03
I'm not really convinced about the 48 MP images, Andrew. It may be that I got a camera that is not so insanely superior. But while I don't find much wrong with the 12 MP images, the result with the 48 MP after no further processing in ACR is not convincing. This shows a 100% crop of the lower right corner of a photo.

[view_image]

Hey Bernd

Sorry magnification is probably the wrong word can you send a small full image to get an idea of how small the crop is?

Also I think pano software might eliminate the edges and stitch only the inner parts of the frame as I have had no issues with 12 MP panos either.

Thanks

Bussty (Andrew)

DAFlys Posted at 9-1 23:45
Good comparison,   I did some shots which included people and the 48mp images certainly showed more details of the people than then 12mp,   Now all I need is an AEB mode for it,  I dont care if its slow.

Yes would be very handy

Bussty Posted at 9-2 02:10
Hey Bernd

Sorry magnification is probably the wrong word can you send a small full image to get an idea of how small the crop is?

»Sorry magnification is probably the wrong word can you send a small full image to get an idea of how small the crop is?«

This is a 970 x 900 px edge crop out of a 8064 x 6048 DNG negative, Andrew. Pure 100% original size!

Can't help, Bernd

Concentrating on perceived detail ignores the main problem with the 48 mode.  Namely the huge colour shifts from debayer on bright, high contrast areas with straight lines.  Road markings go entirely purple and so on.  

SPLflyer Posted at 9-2 02:18
»Sorry magnification is probably the wrong word can you send a small full image to get an idea of how small the crop is?«

This is a 970 x 900 px edge crop out of a 8064 x 6048 DNG negative, Andrew. Pure 100% original size!

Ah thanks Bernd I can scale exactly with the pixel count. So here is that same scene at the corner 970 x 900...


and here it is through Topaz Denoise same size...



Is quite remarkable what Topaz Denoise can do to these images. How do you think the first one compares to your corner crop in terms of image quality?

Cheers

Bussty

This is the main issue with the 48mp bayer processing.  Extra detail is debatable (i doubt it) but the hard to fix bayer artefacts can ruin certain images.  As its not normal CA you cant just fix it easily in post.



Anything with speckle on water below is a major issue too.

Bussty Posted at 9-2 02:41
Ah thanks Bernd I can scale exactly with the pixel count. So here is that same scene at the corner 970 x 900...
[view_image]

That's very difficult to say, Andrew. Your 'corner' has considerably less detail than mine, is closer to the subject and I don't dare say anything about the influence of a possible parallax angle. Because of the lack of detail, the blurriness isn't that noticeable either, but it's there and the massive color noise can't be overlooked either. However, gnirtS' objection cannot be dismissed out of hand either. Unfortunately, none of our photos has road markings in the image ;-).

However, it is indeed amazing what DeNoise AI is still able to get out of the photo. The automatic sharpening applied by DeNoise also makes a positive difference. You are certainly right, with a 48 MP processing to a panorama, the edge areas would be calculated out. But as long as this is not done automatically, I prefer to stay with 12 MB ;-). As AVsupport has explained here https://forum.dji.com/forum.php? ... 251&pid=2841334 , more pixels usually make little sense.

Cheers, Bernd

P.S. gnirtS has just added his example. Something like this is definitely not tolerable!

gnirtS Posted at 9-2 02:46
This is the main issue with the 48mp bayer processing.  Extra detail is debatable (i doubt it) but the hard to fix bayer artefacts can ruin certain images.  As its not normal CA you cant just fix it easily in post.

[view_image]

Hi GnirtS

What's the date of manufacture of your drone? I have never had it this bad on 48MP images. Mine's May 2022. Also is this taken with earlier firmware and is this the DJI jpeg or Conversion from DNG. If conversion did you use Adobe Camera RAW?  One of the few updated to process the Mini 3 Pro DNGs.

Cheers

Bussty

DJI Stephen Posted at 9-1 21:24
Hello there Bussty. Good day and thank you for sharing these interesting information and photos with us. Again, thank you for your support and have a nice day.

sorry if i am rude but apart from social interaction it really  doesn't offer any help about the OP's topic. but then again: neither does my post here. so draw it is

gnirtS Posted at 9-2 02:29
Concentrating on perceived detail ignores the main problem with the 48 mode.  Namely the huge colour shifts from debayer on bright, high contrast areas with straight lines.  Road markings go entirely purple and so on.

Hi Gnirts

Check out these pics 48MP with road markings and very regular close lines. I have never seen that purple fringing so its curious. These were jpegs not from DNG's (forgot to set that setting!)

https://forum.dji.com/forum.php?mod=viewthread&tid=270699

https://forum.dji.com/forum.php?mod=viewthread&tid=271491

I just haven't had your experience...

Cheers

Bussty

tazztone Posted at 9-2 03:09
sorry if i am rude but apart from social interaction it really  doesn't offer any help about the OP's topic. but then again: neither does my post here. so draw it is

Fastest gun draw then surrender in history

SPLflyer Posted at 9-2 03:06
That's very difficult to say, Andrew. Your 'corner' has considerably less detail than mine, is closer to the subject and I don't dare say anything about the influence of a possible parallax angle. Because of the lack of detail, the blurriness isn't that noticeable either, but it's there and the massive color noise can't be overlooked either. However, gnirtS' objection cannot be dismissed out of hand either. Unfortunately, none of our photos has road markings in the image ;-).

However, it is indeed amazing what DeNoise AI is still able to get out of the photo. The automatic sharpening applied by DeNoise also makes a positive difference. You are certainly right, with a 48 MP processing to a panorama, the edge areas would be calculated out. But as long as this is not done automatically, I prefer to stay with 12 MB ;-). As AVsupport has explained here https://forum.dji.com/forum.php?mod=redirect&goto=findpost&ptid=273251&pid=2841334 , more pixels usually make little sense.

Definitely happy with 12MP for Pano's but single shots from the air need all the help you can get from  such a small sensor

Bussty Posted at 9-2 02:11
Yes would be very handy

I have found one issue though with the 48mp images,  they seem to hang the Alamy upload process, I have to upload those via ftp.

gnirtS Posted at 9-2 02:46
This is the main issue with the 48mp bayer processing.  Extra detail is debatable (i doubt it) but the hard to fix bayer artefacts can ruin certain images.  As its not normal CA you cant just fix it easily in post.

[view_image]

I think my drone might be the opposite of your drone, if you check out the 12 / 48 comparison above there are move colour artifacts in the 12MP than the 48MP

DAFlys Posted at 9-2 03:33
I have found one issue though with the 48mp images,  they seem to hang the Alamy upload process, I have to upload those via ftp.

Is that because they are BIG?  Do you sometimes find the odd 48MP or cold even be the first one you take seems to take longer to process until you can take the next one? Just noticed that today...

Bussty Posted at 9-2 03:35
I think my drone might be the opposite of your drone, if you check out the 12 / 48 comparison above there are move colour artifacts in the 12MP than the 48MP

There aren't.  You just dont have the high contrast bright edges where the algorithm goes insane on.

Bussty Posted at 9-2 03:48
Is that because they are BIG?  Do you sometimes find the odd 48MP or cold even be the first one you take seems to take longer to process until you can take the next one? Just noticed that today...

Creation time will vary depending on scene.
48mp isnt 48mpixel.  Its 12mp running through a bayer algorithm to produce a bigger image.  This will take varying amounts of time depending on the scene composition.  Also creating a jpg will do the same.

gnirtS Posted at 9-2 04:19
Creation time will vary depending on scene.
48mp isnt 48mpixel.  Its 12mp running through a bayer algorithm to produce a bigger image.  This will take varying amounts of time depending on the scene composition.  Also creating a jpg will do the same.

Good to know!

gnirtS Posted at 9-2 04:17
There aren't.  You just dont have the high contrast bright edges where the algorithm goes insane on.

I have taken quite a few pics using 48MP and haven't struck this. So you are saying occasionally if you have really high contrast edges you'll get weird colour shifts using 48MP mode?  

What do you make of the colour artifacts in the 12MP comparison above that don't show in the 48MP version. How is that working?

Cheers

Bussty

My regular camera is a Sony A7R2 which is 42MP, it makes the 12 and 48 MP results of the Mini 3 look like utter crap. Which I knew it would before buying the Mini 3.
I just use12MP photos. The 48MP photo feel like I'm just enlarging a poor quality image to create a so called 48MP photo "in camera".
Shooting 12MP, then enlarging it later, I get better results and don't waste space on the SD Card or when backing up the DNG files later on.

SPLflyer Posted at 9-1 22:03
I'm not really convinced about the 48 MP images, Andrew. It may be that I got a camera that is not so insanely superior. But while I don't find much wrong with the 12 MP images, the result with the 48 MP after no further processing in ACR is not convincing. This shows a 100% crop of the lower right corner of a photo.

[view_image]

Mine looks similar, on trees especially looking kind of like an oil painting sometimes.

Hello all

OK so it seems some are getting just as good results resizing 12MP files to 48MP using something like Topaz Gigapixel which I agree is an amazing bit of kit. So I have tried to objectively test this. Before I go on I am just as keen to shoot 12MP to get just as good results as 48MP so I'm doing all I can at all stages to get the best quality on both sides...

So firstly we have a direct 1:1 comparison of the 48MP and the 12MP...



...as just a straight up comparison I think you would have to agree the 48MP has more detail and is clearer plus has a lot less of those weird artifacts between black lines or the blue smudges.

Ok so lets compare the 12MP file if we put it through a well known upsizer Topaz Gigapixel and crank it 2x to match the 48MP image...



...ok this surprised me a little the upsampled 12MP image is pretty close, it's lost a lot of the artifacts and you would almost say it is close enough, this is a 300% crop so that's impressive, you wouldn't normally be viewing at this size unless going big on a wall.  I did notice the upsampling has turned some of those round edges a bit wavy but overall pretty good

But to be fair this is just a table in the centre so lets look at some other parts of the image with this comparison taken from the top left hand corner...



...ok this is a revelation, here is that famous noise we have heard about in the 48MP mode but do remember Gigapixel will reduce noise when it resizes. But clearly the Gigapixel version has less noise but I think it's fair to say the 48MP has lots more detail and is clearer

And another viewpoint with lots of detail and texture...



...OK noise not so noticeable here on the 48MP but what is a standout is the capture by the 48MP of the weave of the wicker. The Gigapixel 12MP upscale has not really caught that.

So the 12MP has been through an improvement process but the 48MP hasn't it seems only fair to compare the 2x Gigapixel results here with the 48MP after it has been through Topaz Denoise AI. Gigapixel does  do some denoising in it's process by default. So here is the 48MP table through denoise compared to 2x Gigapixel treatment of the 12MP file...



...not a massive change for the 48MP image this part of the image with lighter tones wasn't a high noise area in the first place but the wavy lines on the 2x do stand out here.

So lets try the corner...



...so noise is gone now on the 48MP and it's pretty clear this image (48MP) is clearer and more detailed. If you showed me these two images and said which would you prefer for your corners I would have to say the 48MP image but it is closer than I thought it would be.


And finally the chair...



...these are also pretty close, noise is not an issue for either but the 48MP again shows more detail and texture in the wicker on the bottom of the chair.

So my verdict would be if you are after small file size and don't want the wait you get with 48MP then 12MP will do the job. If you are taking very high contrast images with fine lines you may want to shoot 12MP though I have not seen this and in fact if you look above you will see a couple of shots that fit that category but had no issues. I always use the overexpose warning and keep my exposure under that so if over exposing a white line  your mileage may vary.  Also if you are getting up in ISO stick with 12MP as 48MP gets very noisy at the higher ISO's

If however you have time and your prints are going to potentially be made large , the 48MP mode with a run through a good Denoiser will give you the best most accurate detail and clarity.

If you have a different experience and are happy to show a comparison between your 12MP and 48MP results I would love to view. If there is a technique that can get better quality out of the sensor I would love to know.

Many thanks

Bussty

MS Posted at 9-2 06:15
My regular camera is a Sony A7R2 which is 42MP, it makes the 12 and 48 MP results of the Mini 3 look like utter crap. Which I knew it would before buying the Mini 3.
I just use12MP photos. The 48MP photo feel like I'm just enlarging a poor quality image to create a so called 48MP photo "in camera".
Shooting 12MP, then enlarging it later, I get better results and don't waste space on the SD Card or when backing up the DNG files later on.

Hey MS

The Sony is a beast and no comparison but you can't fly it in the sky

I would love to hear your views on my latest post here as I can't get 12MP to upsample as good as 48MP they come close but I couldn't say better. Do you have any examples of 12MP / 48MP or technique you could share? I would seriously love to just shoot 12MP but can't match 48MP quality.

Many thanks

Bussty

Thanks for the in-depth analysis @Bussty, very interesting!
My guess is a post flight considered image processing path with good software and time spent is always gonna have the upper hand than something quickly processed in-camera.
However, currently we are still lacking DNG embedded manufacturer's lens correction profiles (this should have been fixed previous FW) which could make all the difference narrowing the gap between what you're currently experiencing. IMO AI driven software always performs better the RAWer and the more of the data is.
I've since made my own LCC in CaptureOnePro22, AI-denoise and can sharpen if required in ON1 but for my output being 360 I always end up downrezzing to 14Kx7K anyway using 12MP to start with. This still gives you 8K VR and creates less hassle with throughput. But that's just me, and I'm not really interested in single images (at the same token I would love +/-3 AEB)

https://kuula.co/share/Nxlts?log ... p;sd=1&thumbs=1
https://kuula.co/share/Nxltz?log ... p;sd=1&thumbs=1

Bussty Posted at 9-2 03:48
Is that because they are BIG?  Do you sometimes find the odd 48MP or cold even be the first one you take seems to take longer to process until you can take the next one? Just noticed that today...

Ive uploaded bigger so I dont know,  they process fine when uploaded by ftp.

AVsupport Posted at 9-2 22:48
Thanks for the in-depth analysis @Bussty, very interesting!
My guess is a post flight considered image processing path with good software and time spent is always gonna have the upper hand than something quickly processed in-camera.
However, currently we are still lacking DNG embedded manufacturer's lens correction profiles (this should have been fixed previous FW) which could make all the difference narrowing the gap between what you're currently experiencing. IMO AI driven software always performs better the RAWer and the more of the data is.

Hey AVsupport
Thank you!   

DXO is imminently about to release a profile so really looking forward to that.

Impressive Panos! That second one really shows what the Mini 3 Pro can do above all other drones and yes agree 12MP for 360 spheres make sense (they are big final files!).

I work in both spaces so keen to work out best process for both. Pretty happy with the 360 sphere process especially since DJI almost made the sphere 100%. So just sitting waiting patiently

Cheers

Bussty

DAFlys Posted at 9-3 00:17
Ive uploaded bigger so I dont know,  they process fine when uploaded by ftp.

That's odd isn't it.

gnirtS Posted at 9-2 04:19
Creation time will vary depending on scene.
48mp isnt 48mpixel.  Its 12mp running through a bayer algorithm to produce a bigger image.  This will take varying amounts of time depending on the scene composition.  Also creating a jpg will do the same.

Have you noticed though (which is weird) the actual capture is at the end of the wait? Have confirmed this by capturing a digital clock with seconds hand. You would think it would all be about post processing and not preprocessing?

Bussty Posted at 9-3 01:40
Have you noticed though (which is weird) the actual capture is at the end of the wait? Have confirmed this by capturing a digital clock with seconds hand. You would think it would all be about post processing and not preprocessing?

Could it be though that the sensor is read left to right so the seconds was in the last part of the capture.  

DAFlys Posted at 9-3 02:07
Could it be though that the sensor is read left to right so the seconds was in the last part of the capture.

I think it was in the middle but makes me want to go outside with a high shutter speed and 48MP mode and spin the drone as  I take the picture as if what you say is correct it should look very odd?? Will try this tomorrow but I didn't think that's how the 48MP capture works. Would be great if anybody can  explain what the pre-capture time is for?

Bussty Posted at 9-3 02:23
I think it was in the middle but makes me want to go outside with a high shutter speed and 48MP mode and spin the drone as  I take the picture as if what you say is correct it should look very odd?? Will try this tomorrow but I didn't think that's how the 48MP capture works. Would be great if anybody can  explain what the pre-capture time is for?

Not related to speed but did you see this review of 12 vs 48mp on the air2s.

DAFlys Posted at 9-3 02:50
Not related to speed but did you see this review of 12 vs 48mp on the air2s.

https://www.youtube.com/watch?v=wR6jo-hKwog

Thanks DAFlys feels like those results match closely with mine together with 48MP having more noise. When you scan through those comparisons above without reading the MP it's pretty easy to tell which one looks better and from a detail point of view it's always the 48MP. I'm not sure why there is such a resistance in using it when it's going to give you more detail?   

Bussty Posted at 9-2 17:35
Hey MS

The Sony is a beast and no comparison but you can't fly it in the sky

I use Topaz Gigapixel too, but I think x2 upsizing doesn't look good, too plasticy at times.
I set it to x1.5. It seems a good compromise in upscaling, without having truly massive TIFF files to upload elsewhere or to backup locally.