Mavic 3 color shifting in white balance / color tint

[JustName]

I don't think there are any changes on the files by PS, since the file was just opened. But here are two files directly from memory card.

Download...

 

[Skyscope]

The metadata is different. I'm interested in comparing the not manipulated metadata between both FW versions myself, even if the images may look the same.

 

[globetrotterdrone]

So, basically, what I expected and DJI is trying to do is fixing some (strange) unwanted optical behaviour with software alteration in the RAW files.

While this is certainly not uncommon as raw data always needs to be processed (debayering, etc.) as @Skyscope pointed out, I still wonder how this just slipped through their QA and how they judge this amount as small or within tolerance. Well, maybe there's no real QA anyway (coming from the soft lens M2P story).

Good luck with the change.

 

[Skyscope]

I still wonder how this just slipped through their QA and how they judge this amount as small or within tolerance.

Sure, there is most probably a spread in this early state of production, like it always was with DJI drones. Think of the massive purple flaring in early Mavic Pros (1), diffraction with first M2P camera modules, etc. This wil be ironed out with time, with new hardware revisions. The Inspire 1 for example had 4 revisions. As another example, I have two M2P, one from 2018, the other from late 2019, both are very different in diffraction at different f-stops.

But those color vignetting / fringing / distortion issues with the Mavic 3 are not there because of a lack in DJI's QA. They are basically inherent to the system. Take a look at regular 1", M43, APSC, Full Frame cameras - the larger the sensor, the bigger and heavier the lenses are. Cunsumers today are demanding bigger sensors, but they usually come at the cost of more complex lenses. Not possible on this drone (or any other in such a small form factor). To full-fill the demand of the market, DJI has to take these compromises, and so does the customer. So it's completely debatable, if a M43 sensor in such a small drone without the possibility to get a proper lens is really a good idea to buy, besides its better sensitivity in low light situations.

 

[mischakeijser]

Sure, there is most probably a spread in this early state of production, like it always was with DJI drones. Think of the massive purple flaring in early Mavic Pros (1), diffraction with first M2P camera modules, etc. This wil be ironed out with time, with new hardware revisions. The Inspire 1 for example had 4 revisions. As another example, I have two M2P, one from 2018, the other from late 2019, both are very different in diffraction at different f-stops.

But those color vignetting / fringing / distortion issues with the Mavic 3 are not there because of a lack in DJI's QA. They are basically inherent to the system. Take a look at regular 1", M43, APSC, Full Frame cameras - the larger the sensor, the bigger and heavier the lenses are. Cunsumers today are demanding bigger sensors, but they usually come at the cost of more complex lenses. Not possible on this drone (or any other in such a small form factor). To full-fill the demand of the market, DJI has to take these compromises, and so does the customer. So it's completely debatable, if a M43 sensor in such a small drone without the possibility to get a proper lens is really a good idea to buy, besides its better sensitivity in low light situations.

But these are no normal lens abbarations, like vignetting / fringing or distortion (which are correctable in post).
This seems like problem with the sensor.

 

[Skyscope]

Very very unlikely.

 

[c.oberschneider]

Put a cheap wide angle lens on your dslr or mirrorless camera, take a shot of a white wall and then crank up dehaze in lightroom et voilà - welcome to greenish magenta wonderland. That's not a result of a defect camera sensor either.

 

[Point Zero]

Is it possible that the reason we have still yet to see Panoramas available on a 3K drone is because of this issue?

I mean imagine what a pano would look like with this issue present in every separate image which is then stitched together and why it was still not made available on the last FW update, even though technically its way less work in terms of some of the features they actually released.

Just a thought...

 

[calmar81]

Maybe you’re right… could be the case.

I've also asked myself why it should be more time-consuming to code some software for the same, relatively simple sequence of gimbal movements and yaw movements of the drone than focus track, POI or even master shots, in which the drone has to carry out multiple, autonomous movements that require „recognizing“ and following an object.

I think coding is not the complex thing here… and I think the movement sequence itself should be finished or adapted from other DJI drones, they just have to change how far gimbal and drone turn between the shots because of changed focal length. Maybe they just have to wait for a fix for optical aberrations from the „optics team“ first?

Maybe we learn more about this soon.

 

[Gibonxxl]

Have any of you thought this colour shifting in white balance might be a problem with some light scattering that is coming in through the top lens?
Perhaps taping the upper zoom lens with tape will answer this question.

 

[flyliedutch]

Have any of you thought this colour shifting in white balance might be a problem with some light scattering that is coming in through the top lens?
Perhaps taping the upper zoom lens with tape will answer this question.

That’s technical not possible. That are 2 lenses with each it’s own sensor.

 

[Coskier]

That’s technical not possible. That are 2 lenses with each it’s own sensor.

I haven't taken the clear filter off to check, but it could be possible if there was any gap between the filter and the lens housing. It's actually an interesting thought to me.

 

[flyliedutch]

I haven't taken the clear filter off to check, but it could be possible if there was any gap between the filter and the lens housing. It's actually an interesting thought to me.

Why would they design a gap between those 2 cameras? Makes no sense to me.

 

[Coskier]

Never said it was designed. But if one gets introduced through vibration or whatever, then it could explain some things. Just saying it is possible.

 

[Skyscope]

It seems like DJI got it right with the current firmware, if your camera doesn't suffer from production spread.

From today with current firmware one Top Down shot of a snow-covered field with overcast sky, so perfectly even lighting conditions.

• The first image is the result from a conversion with Raw Therapee, showing the true quality of the lens withoutany modifications. It show the usual vignetting from these kind of compact lenses (see Post #224).
• The second image got a Flat Field correction with Raw Therapee only, nothing else. No more color vignetting.
• The third image was converted with Lightroom, with Lightrooms standard application of Opcodes embedded in the DNG. No further settings applied.
  You will notice, that besides an automatic boost in exposure, contrast and saturation it gets a distortion correction with a crop too. The resolution of the image container stays at 5272 × 3948, so all images out of Lightroom are slightly upscaled.

(For best comparison, click on one image on your Mac/PC and switch between them with your cursor keys)



Regarding video, I got comparable results with D-Log Clips from the same Mavic 3: No vignetting or color shifts.

 

[Kimblah]

It seems like DJI got it right with the current firmware, if your camera doesn't suffer from production spread.

From today with current firmware one Top Down shot of a snow-covered field with overcast sky, so perfectly even lighting conditions.

• The first image is the result from a conversion with Raw Therapee, showing the true quality of the lens withoutany modifications. It show the usual vignetting from these kind of compact lenses (see Post #224).
• The second image got a Flat Field correction with Raw Therapee only, nothing else. No more color vignetting.
• The third image was converted with Lightroom, with Lightrooms standard application of Opcodes embedded in the DNG. No further settings applied.
  You will notice, that besides an automatic boost in exposure, contrast and saturation it gets a distortion correction with a crop too. The resolution of the image container stays at 5272 × 3948, so all images out of Lightroom are slightly upscaled.

(For best comparison, click on one image on your Mac/PC and switch between them with your cursor keys)

View attachment 140414View attachment 140415View attachment 140416

Regarding video, I got comparable results with D-Log Clips from the same Mavic 3: No vignetting or color shifts.

So the conclusion is that it is a lottery to get a properly working unit. My fourth Mavic 3 should arrive on Tuesday and if it still suffers from this color shift then I'm done with Mavic 3 until production quality is on a better level.

 

[JustName]

So the conclusion is that it is a lottery to get a properly working unit. My fourth Mavic 3 should arrive on Tuesday and if it still suffers from this color shift then I'm done with Mavic 3 until production quality is on a better level.

No.

I have seen a lot of JPGs and DNGs made by a Mavic 3 in the last days.

All of them had the same issue. The only difference is only the image itself, in some images the issue is more visible, in others less.

Also the example from the last post. Just take the image (in best case the dng and not the manipulated jpg) and set dehaze, etc. on a high value. You will see almost the same picture as on the examples bevor.

 

[Skyscope]

Also the example from the last post. Just take the image (in best case the dng and not the manipulated jpg) and set dehaze, etc. on a high value. You will see almost the same picture as on the examples bevor.

I don't get the point here.
Nobody would dehaze or work in any other way with such an image before applying any corrections.

Like I wrote several times, this sort of vignetting isn't unusal for lenses in compact camera modules. Throw RAW images of a Mavic 2 Pro, Air 2S or even from a smartphone in RAW Therapee, and you see similar results. Yes, it's worse on the Mavic 3, but because of it's bigger sensor, which would need a much much bigger lens than the other drones with their smaller sensors, to be comparable in optical regards.

To sum it up, unevenly color smearing or spots are problems, probably because of a decentered lens.
But this sort of vignetting isn't.

 

[JustName]

I don't get the point here.

I am really sorry for you.

Nobody would dehaze or work in any other way with such an image before applying any corrections.

No one said that you would, or should do so.

Like I wrote several times, this sort of vignetting isn't unusal for lenses in compact camera modules. Throw RAW images of a Mavic 2 Pro, Air 2S or even from a smartphone in RAW Therapee, and you see similar results. Yes, it's worse on the Mavic 3, but because of it's bigger sensor, which would need a much much bigger lens than the other drones with their smaller sensors, to be comparable in optical regards.

Sure, therefore just take a DNG-File from a random M3, do the same manipulations like you did it in the post #233 and you will get almost the same result. There is nothing special on the M3 you did used for this pic.

To sum it up, unevenly color smearing or spots are problems, probably because of a decentered lens.

So much decentered lens will be seen much earlier on the decentered focus.

 

[Paradigm]

I haven't noticed this issue with my M3.

Is there a specific test I can do to test this out? What I mean is, taking a photo of a white wall and then applying filter/effect XYZ on the .dng file...

Something of that nature spelled out step by step so we can all test our M3's out and perhaps we can come to a better understanding just how widespread this issue is. Maybe reference it with production date and general location of purchase.

Thoughts?