Which picture ratio?

[andy_k]

Ok- let's consider the OP's original question. The 3:2 image will, to use his words, give the fullest 20mp image. In fact it is the only available aspect mode that can produce a 20mp image. I don't think anyone is disputing that the 16:9 image seems to be a tiny bit wider across the frame. To speak for myself (although I would expect most who have even a casual interest in photography would agree) I am not discounting it I am saying it's of no consequence. To fly the additional altitude of distance required to make up for it is a trivial excersize.

If the 16:9 is a 'tiny bit wider' that means that the 3:2 is not using the full sensor doesn't it?

If it's not using the full sensor it cannot ever produce a genuine 20mp image

As we are getting 20mp DNG files something is clearly not right - see my post number 10 which shows what the Phantom captures 20mp and what it delivers - in this case approximately 15mp which means roughly 25% of the original recorded data has been discarded.

The final image has been distorted to make it look good but this means that things go out of position (rough illustration in post 30) so aligning panoramas is not as easy as it should be.

What is also contributing to the final image issues people are seeing it that we are presented with a 15mp image being enlarged to 20mp in camera.

I mentioned the difference in results from the two different formats a couple of times - if we gain a couple of cms at a few metres range that has to increase as we get to larger distances - as you say, the field of view determines that. If it were constant then it would be easy to compensate but as it appears to be decided 'in camera' it's a variable that a people didn't expect and some of us don't really want to have to deal with.

 

[andy_k]

Any lens corrections you might apply in pre processing should be a simple click once you have the profile.

Yep, it should but this is one of the things that first flagged up that there was something 'not quite right' - the required profile changes because the output from the camera is not constant

 

[With The Birds]

If the 16:9 is a 'tiny bit wider' that means that the 3:2 is not using the full sensor doesn't it?

If it's not using the full sensor it cannot ever produce a genuine 20mp image

As we are getting 20mp DNG files something is clearly not right - see my post number 10 which shows what the Phantom captures 20mp and what it delivers - in this case approximately 15mp which means roughly 25% of the original recorded data has been discarded.

The final image has been distorted to make it look good but this means that things go out of position (rough illustration in post 30) so aligning panoramas is not as easy as it should be.

What is also contributing to the final image issues people are seeing it that we are presented with a 15mp image being enlarged to 20mp in camera.

I mentioned the difference in results from the two different formats a couple of times - if we gain a couple of cms at a few metres range that has to increase as we get to larger distances - as you say, the field of view determines that. If it were constant then it would be easy to compensate but as it appears to be decided 'in camera' it's a variable that a people didn't expect and some of us don't really want to have to deal with.

To the extent the 3:2 image depicts a narrower FOV across the frame than one shot at 16:9 then yes it seems some of the sensor readout at the extreme edges is discarded during the in camera processing applied.

All lenses, especially those of wide angle design, exhibit some form of distortion (almost always most pronounced towards the edges of the frame) and light fall off. I think we might all agree that it is disappointing to what extent these issues are present in the P4P.

Rather than grizzle and complain we probably should be happy it is as good as it is (there are to my knowledge no competitor products that come close to it at the price point) and look forward to the phantom 5 or whatever the next improved model might be.

The issues with stitching can be easily resolved by providing for a more generous overlap at the time of image acquisition. With a little care in post we are able to produce ridiculously good images having regard to the money we outlay on this gear.

I must confess I might be missing your point on the FOV somehow having a greater potential effect on distant objects. If the lens design/quality provides that the level of distortion is unacceptable at and beyond some point out from the centre of and continuing towards the edge of the frame it simply determines a percentage of the frame that might be discarded in processing. It's a fixed percentage regardless of what the subject is or how far away from the camera it might be (assuming acceptable depth of focus).

 

[AyeYo]

Meta,

I don't think many of us are concerned with the FOV on the uncorrected, true RAWs, because they're a hassle to work with. The facts seem to show that the correction applied DNGs (which is what 99.999% of P4P users are working with) yield larger FOV with 16:9.


The panorama stitching issues are major and they're actually part of what prompted me to ask this question. Even with massive overlap (and it's 16:9, 4:3, or 3:2 make no difference), I cannot get PTgui to give me anything better than "not so good" and often visually evident stitching issues with the P4P. While I can shoot Mavic panoramas with absolutely minimal overlap in 16:9 or 4:3 and they all stitch great.

Now I'm beginning to understand why I'm not seeing significant differences in the P4's dng vs. jpg when I try to get the most out of my images. In fact, I think I'm just fooling myself and have been taking it on faith that the dng would yield better details. Those who have been touting working with dng as advantageous I suspect are just touting the party line with no real verifiable results.

My Nikon nef on the other hand yields tremendous results as compared to its jpg when it comes to revealing details and latitude.

I now plan on using UFRAW to see what I can see.

There isn't an unedited difference but you absolutely have better editing leeway with the DNGs. Don't give up on them unnecessarily.

 

[AyeYo]

I must confess I might be missing your point on the FOV somehow having a greater potential effect on distant objects. If the lens design/quality provides that the level of distortion is unacceptable at and beyond some point out from the centre of and continuing towards the edge of the frame it simply determines a percentage of the frame that might be discarded in processing. It's a fixed percentage regardless of what the subject is or how far away from the camera it might be (assuming acceptable depth of focus).

It's a fixed percentage *of the frame size*, but as the shot gets wider and wider (or more distant, however you want to look at it), that translates to more and more feet on the ground. So while you might only be losing inches when doing test shots off your kitchen table, you could be losing entire buildings/mountains/etc. that are a mile away when shooting a landscape.

 

[AyeYo]

Whether its a distant mountain range or a tree your hovering 10m behind makes no difference and you will see the same proportionate amount more or less of each with any incremental change of FOV.

The issue isn't that the images are cropped. It's that they're cropped and then rescaled back to the proper resolution. That's what causing the stitching errors and why simply overlapping more doesn't fix it.

 

[With The Birds]

It's a fixed percentage *of the frame size*, but as the shot gets wider and wider (or more distant, however you want to look at it), that translates to more and more feet on the ground. So while you might only be losing inches when doing test shots off your kitchen table, you could be losing entire buildings/mountains/etc. that are a mile away when shooting a landscape.

Yes, no question. It should have no effect on what the correction profile for the lens is though.

 

[With The Birds]

The issue isn't that the images are cropped. It's that they're cropped and then rescaled back to the proper resolution. That's what causing the stitching errors and why simply overlapping more doesn't fix it.

This might only be an issue if the crop and resize applied to one dimension only (i.e. lop 150 pixels off each side then use some form of algorithm to make the resultant crop x+300). If the whole frame was cropped to constrain dimension and upsized it should cause no problems. The correction applied is seemingly more complex than cropping and resizing.

 

[KevMo Photog]

Guys I have read all of this thread for a few days now! Like I read one guy say we should be pretty happy at the results we are getting from this tiny sensor and lense! They will continue to tweak sensor and lense combo's as we go on this journey. They have a lot to deal with when trying to make everything line up with this small sensor and lense. Stitching is fun and very cool to do but it's not ever gonna be as perfect as our full frame DSLR's produce! So I would suggest like has been said, to overlap a little more if your going to stitch images together. Fun watching you guys spar over these little tiny pixels though! LOL

 

[AyeYo]

It's not about tiny pixels. It's about DJI advertising one thing and delivering another, then not being up-front about it. No one is complaining about image quality. We're simply trying to figure out wtf is going on, since DJI didn't bother to be honest about it and spell it out in the manual like they should have.

 

[andy_k]

yes it seems some of the sensor readout at the extreme edges is discarded during the in camera processing applied.
.

it's not the 'extreme edges - it's about 20% of the image - check my post where I showed the two different output

The issues with stitching can be easily resolved by providing for a more generous overlap at the time of image acquisition.

Not quite so simple, the correction algorithm isn't just a crop, it's scaling/warping parts of the image (again see my previous posts) so when it comes to stitching, even the middle areas of the frame have some distortiion. On expansive landscapes it doesn't matter, on shots where there are lots of string verticals, it's easily sorted but in complicated urban landscapes a lot of manual correction is needed if you don't want duplicate houses etc in the final panorama

I must confess I might be missing your point on the FOV somehow having a greater potential effect on distant objects. If the lens design/quality provides that the level of distortion is unacceptable at and beyond some point out from the centre of and continuing towards the edge of the frame it simply determines a percentage of the frame that might be discarded in processing. It's a fixed percentage regardless of what the subject is or how far away from the camera it might be (assuming acceptable depth of focus).

OK, let me try and explain what I'm saying

As (some of us seem to think) the 16:9 utilises more of the original raw data (not the DNG) and that it sometimes renders a wider view in the same final format it effectively has a (albeit slightly) wider field of view.

Close to the camera that could result in images just a couple of centimeters wider but as you move further away from the camera that gap between the two get's much wider - at 3m a couple of centimeters could easily become dozens of meters as your target moves further away from the camera.

Because the camera is correcting images prior to saving them it appears that it's not 100% accurate and we do get some variation in the perceived focal length

 

[andy_k]

This might only be an issue if the crop and resize applied to one dimension only (i.e. lop 150 pixels off each side then use some form of algorithm to make the resultant crop x+300). If the whole frame was cropped to constrain dimension and upsized it should cause no problems. The correction applied is seemingly more complex than cropping and resizing.

the crop from the original raw and subsequent scaling is very definitely degrading the image .....see below

left hand side is a 100% crop direct from an unedited DNG file with RawTherapee which disregards whatever profile DJI apply to their raws
right hand side is a 100% crop from the same unedited raw file in Adobe Camera raw which does apply DJIs embedded profile

Both shots show the extreme left hand edge of the frame without any edits or tweaks

The left hand one (RT) shows far more detail - look at the timber decking and compare the size of the person in the black suit - he's 30% taller in the DJI 'version', the flags are sharper, the huts are better defined etc etc. They've all been enlarged when compared to 'what the camera saw' and that makes any further edits difficult

What is more of an issue is the clear distortion at the edge of the right hand image - everything is stretched horizontally and just appears 'muddy' when compared to the left hand image.

I

 

[andy_k]

another clip from the same original shots - this should (hopefully) illustrate the distortion/correction that is being applied to our 'raw' files when we get the DNGs

Note the position of the boarder in relation to the surf on the wave in the left hand image, then check it on the right! These are both taken from exactly the same file. The images are being squeezed in at the centre to correct barrel distortion.

Again...look at the lack of sharp detail on the right and how much larger (and wider) the only remaining boarder is - this is the extreme bottom left of the same image as the previous post

 

[AyeYo]

Those are really great examples.

So I guess the next step is... is there a way to Lightroom import without applying the built in profile? I don't see anything jumping out at me, as the profile appears to be getting applied in the background rather than in the typical lenses correction section. Do you have to import to a non-Adobe program to see the actual RAW picture? I'd be perfectly happy applying my own cropping and distortion corrections.

 

[andy_k]

RawTherapee Features

Rawtherapee is free and incredibly powerful - process there rather than Adobe Camera Raw/Lightroom and export as png

 

[AyeYo]

RawTherapee Features

Rawtherapee is free and incredibly powerful - process there rather than Adobe Camera Raw/Lightroom and export as png

I'll use that if i have to, but it'd be nice to avoid an extra step in the post-production process if possible.

 

[Meta4]

The panorama stitching issues are major and they're actually part of what prompted me to ask this question. Even with massive overlap (and it's 16:9, 4:3, or 3:2 make no difference), I cannot get PTgui to give me anything better than "not so good" and often visually evident stitching issues with the P4P. While I can shoot Mavic panoramas with absolutely minimal overlap in 16:9 or 4:3 and they all stitch great.

Stitching issues?
About half of my shooting is panoramas and I haven't encountered these stitching issues.
This example is stitched from 10 individual images.
Perhaps you should try another stitching program?

 

[andy_k]

Stitching issues?
About half of my shooting is panoramas and I haven't encountered these stitching issues.
This example is stitched from 10 individual images.
Perhaps you should try another stitching program?

As I said earlier in the thread

Not quite so simple, the correction algorithm isn't just a crop, it's scaling/warping parts of the image (again see my previous posts) so when it comes to stitching, even the middle areas of the frame have some distortion. On expansive landscapes it doesn't matter, on shots where there are lots of string verticals, it's easily sorted but in complicated urban landscapes a lot of manual correction is needed if you don't want duplicate houses etc in the final panorama

Now, unless I'm very much mistaken - that example you posted falls into the 'expansive landscape' category

 

[DefiantChild]

Stitching issues?
About half of my shooting is panoramas and I haven't encountered these stitching issues.
This example is stitched from 10 individual images.
Perhaps you should try another stitching program?

Maybe that's because you're only pano stitching 10 images. Try orthomosaically stitching 500 or 5,000 aerial images (which is what, I believe, FunN4lo was attempting to accomplish for his client [albeit with 15 images]).

 

[Meta4]

Maybe that's because you're only pano stitching 10 images. Try orthomosaically stitching 500 or 5,000 aerial images (which is what, I believe, FunN4lo was attempting to accomplish for his client).

And I have no problems stitching big orthophoto mosaics either.
This one is 128 images: