Air 3 vs DSLR Experiment - Help needed again... Thank you!! (First result in...)

[Felix le Chat]

Thank you so much. I have been completely baffled about all this negativity and I really need to ignore it and just show the data I receive. To me while stitching has been around a long time on drones to be able to stitch a tele lens on an Air model to surpass the quality of a single frame will improve my stills taken from the air enormously. Maybe some just don't stitch or aren't familiar with it but I use it not just with drone images but also my Mirrorless system and is very familiar to me.

If you want a way to keep drones small but give them the ability to capture scenes in true high res, this is it. Why would you not want to explore that possibility especially when evidence supplied clearly shows a marked improvement.

I appreciate your viewpoint on this.

I understand your intent, the Mavic 2 zoom had something similar baked into one of the firmware updates, they named it 'super-resolution'. The camera used optical zoom to 48mm equivalent and shot a 3x3 grid pano which was stitched in-camera. The results were extremely good. Detail being greatly enhanced, so maybe DJI can do a bit of archive digging to re-purpose existing code for the Air 3 70mm camera.

 

[SkyeHigh]

Thanks again to noob1957 on the DJI forum for going out of their way to test the Air 3 against their Sony FF 24MP camera.

So to recap noob1957 has found a couple of scenes in midday sun and taken wide 24mm views with both the Air 3 and a Sony FF DSLR. Here are those views…

View attachment 169391

View attachment 169392



The whole feature of this technique is to then stitch a 16x 70mm shot grid taken from the same spot as the 24mm images to effective cover the same area as the 24mm lens. We aren’t doing that in this test as we have tested the ability to stitch those images here… https://forum.dji.com/forum.php? ... D922%26typeid%3D922 and it works really well as can be seen (remember these images are just jpegs so properly converted RAWS could be even better)

So these next two comparisons are just a zoom of a single central image from the Air 3 Tele lens compared to a single frame of the Sony FF 24MP camera taken on a bright day. The Sony image is zoomed in to match the 70mm Tele of the Air 3. What we are really testing here is if we could send our Sony FF 24MP DSLR flying into the sky are we better capturing a single 24mm Wide Angle shot with the Sony DSLR or 16x Air 3 images using the 70mm lens then stitching to provide the same 24mm view as the Sony. You will have to agree that in both cases the Air 3 70mm result is significantly better…

View attachment 169393

View attachment 169394


And that result means with a little work in post processing to stitch 16x 70mm images (honestly takes less than 2 mins) you will exceed the quality of most DSLR cameras and any Micro 4/3 camera found on the Mavic 3 Pro or Mavic 3. And that fact is revolutionary for stills photographers especially if DJI make this automated 16x capture a quick and seamless process which they absolutely can do if they want to. Of course this won’t work if your image is moving but for the majority of still photographers that is not the case and with stitching software today and the ability to mask you can easily get around this for small areas of motion in your image.
You can download the source material for this latest test and see for yourself https://drive.google.com/drive/f ... bilY?usp=share_link

Thanks again for noob1957 taking the time to conduct the test and if this technique interests you please let DJI know including voting here if you haven’t already…

https://forum.dji.com/forum.php? ... D922%26typeid%3D922

Currently 44 of you think this is a great idea

Just out of interest, why are the Sony shots being taken at so large an aperture? At those apertures you’ll never get the best from what a DSLR lens can produce. SLR/DSLR lenses are designed for peak performance around f8.0.

 

[Bussty]

I understand your intent, the Mavic 2 zoom had something similar baked into one of the firmware updates, they named it 'super-resolution'. The camera used optical zoom to 48mm equivalent and shot a 3x3 grid pano which was stitched in-camera. The results were extremely good. Detail being greatly enhanced, so maybe DJI can do a bit of archive digging to re-purpose existing code for the Air 3 70mm camera.

Hi Felix Yes that sounds like a very similar process looks like a case of back to the future needed by DJI

 

[Bussty]

Just out of interest, why are the Sony shots being taken at so large an aperture? At those apertures you’ll never get the best from what a DSLR lens can produce. SLR/DSLR lenses are designed for peak performance around f8.0.

Hi SkyeHigh until we can fly DSLR's around it's probably not a fair comparison to test against optimal aperture I'm not sure what the optimal fstop for the Air3 sensor would be on the 70mm so I think having the two as close as relatively possible is a good comparison. Th only tool we really have is the lens on the drone so while interesting the DSLR comparison is kind of theoretical the real comparison is comparing the drone single frame to the 16x 70mm stitch and as we have seen that is night and day.

 

[landscapetog]

The camera and lens in any consumer/pro-sumer drone will never match a DSLR/Mirrorless camera. The image sensor is substantially smaller, less sensitive and noisier and the lens (often plastic) will never be as sharp as that of a DSLR/Mirrorless camera with a real glass lens or such good control of distortion, chromatic aberration, noise…

Trying to stitch an image with so many frames will rarely correct for all the distortion in each shot.

The IQ from the 12mm lens on my M3 is not equal to a single frame from my A7R4, A1 etc, but it comes close in most respects when used in bracket mode and merged to HDR.

Far easier to bracket / merge images from the M3 than from a handheld dSLR or mirrorless camera, because the gimbals are extremely effective.

With stitching + auto bracket, the M3 images from the wide camera easily rival single frames from my Sony 50 and 60 MP FF cameras. And it's far easier to stitch from the drone than from handheld use of my mirrorless bodies.

Of course there is no comparison in the angles / POVs, far easier to compose the image you want with the Mavic than the A7R4.

Both systems are indispensable, but I'd take the Mavic over the Sony FF system if I could only keep one.

 

[landscapetog]

Hey there firsthobby

I own a Mini 1 and Mini 3 Pro which I believe has a similar sensor to the Air 3 but you are right I don't own an Air 3 and I haven't personally had the experience of a tele lens on a drone yet, however I have had over 3 years experience with drones and it has become clear to me that especially down at the Mini and even to an extent the Mavic level there is a difference in quality in stills compared to a DSLR camera and that is what I'm trying to address. Luckily existing owners of the Air 3 have helped me collect real data on this idea to prove it is very possible.

If we can convince DJI to automate the process it will take about as much time to take as a Wide Angle Pano and way less time than a full 360 pano.

I really think if drone photographers know they can get quality as good or better than a DSLR (even approaching a DSLR) they will be more than happy to stitch 16 images, using PanoramaStudio3 it takes less than a minute to stitch. For big prints this technique delivers outstanding detail which you just can not achieve with a single wide shot.

I look at all the ideas started in these forums in the past... Waypoints for the Mini, 360 avoidance for the Mini, effectively return of the tripod mode across most drones, 360 stitching on the Mavic 3 using the tele lens and these all have become reality just like I'm hoping the idea of effortlessly capturing a high res image of a 24mm scene will become a reality also.

Cheers

Once I figured out the 12mm lens (24mm equiv) could provide something close to my FF Sony cameras in 5-shot bracket mode + HDR, that became a complete game changer.

Using my drones far more and the mirrorless camera far less these days.

Automating the 70mm equivalent cameras to auto-capture the area of the 24mm camera (particularly in combination with bracketing) would be a huge win for DJI drones.

 

[davidarmenb]

I’m going to jump into this wagon and do some tests as well!

I own a 32.5MP CANON EOS 90D Crop sensor DSLR and will be purchasing the Air 3 within the following week.

I will take the DSLR test shots using a basic Canon EF-S 10-18mm lens at 15mm with a variety of apertures.

 

[MaSz]

Where this breaks down is that that a relatively new FF camera will provide you with much greater dynamic range, reduced noise, and wider latitude for post-processing than the tiny sensor in an Air3, so it's not really an apples-to-apples comparison and raw megapixels don't tell the whole story. Also, at some point, you just aren't going to see those extra megapixels unless you are peeping or making huge prints.

This!
I don't own air3 yet and I'm surprised that there's no pano option for tele lens!
You want beat Mirrorless camera with air3 couse Raw files on Air 3 are 10bit? You can't push them in postprocessing as far as 14bit raw files from big FF camera sensor. But what you'll get with zoom pano is completely different perspective and thats what I'll be looking for with zoom panos.

 

[GadgetGuy]

I'm surprised that there's no pano option for tele lens!
But what you'll get with zoom pano is completely different perspective and thats what I'll be looking for with zoom panos.

There's always the manual pano option, which will be far more efficient than the current automated 70mm pano option on the Mavic 3 Pro, which takes 4.5 minutes and still doesn't stitch the 128 images, of which the top row is unusable because it is detail-less sky. Shooting only 96 images will be faster and stitch properly. Use the grid overlay for 1/3 overlap between images.