Lens corrected...

Crop of lens correction (all video and images are derived from this final crop - video is about 1:1 or down scaled and images are up scaled back to 20MP):

One important thing to note is that in the examples and when using a lens correction tool that bends inwards like in the photos then up scaling appears the end result to get back to 20MP. However its not really up scaled in this way. Using a proper lens profile it is bent (if you like) the other way (corners are pulled outwards rather than edges pushed inwards. The end result is still some stretching / up scaling of the outer edges along with some cropping.

Dingoz Posted at 2017-9-27 20:31
For those interested this is how the video and images are processed into final outputs (approximately):

DJI R&D said they use the entire width of the sensor to capture video. in your diagram it is just a center crop.

EDIT: ah sorry, in understand now - the area marked as video is the final resulting video area after full width readout + correction + cropping, right?

Dingoz Posted at 2017-9-27 20:33
Original 20MP sensor + lens output (Raw)

i have a very hard time believing the uncorrected lens is this bad.  you re saying if i take a RAW photo 3:2 5472×3648 and not do any lens correction I would end up with severe fisheye effect and like -4EV vignetting?

i am assuming the RAW comes out uncorrected, right? (otherwise how would you get these charts)

EDIT: holy crap, i actually went home during lunchbreak and tested it - you are correct. the lens is pityful (when compared to what I am used to with quality photo lenses), and correction is incredibly destructive.. and to make matters worse they even scale back up to 20MP after distorting the crap out of it.. why?
hard to believe it ends up being as good as it is with this sorry piece of plastic as a lens

well, a couple more tests show

- OnePlus3 cellphone lens actually covers sensor (1 2/3" sensor) much better, similar to what we d see from DSLR lenses
- RX100m4 at 24mm has similar poor converage / high distortion as P4pro (both 1" sensor). I did not expect this. However, this is a compact zoom lens whereas P4pro is fixed.

So I assume the distortion of P4Pro lens isnt that unusual. Still, a bit disappointing/unexpected what level of correction is applied to ALL footage we get from that camera

It s like taking a full frame sensor, then adding an APSC lens, cropping to APSC coverage, then upscaling and saying Taddaaaa FullFrame Camera!!
;)

Sebb Posted at 2017-10-2 01:10
DJI R&D said they use the entire width of the sensor to capture video. in your diagram it is just a center crop.

EDIT: ah sorry, in understand now - the area marked as video is the final resulting video area after full width readout + correction + cropping, right?

This might help clarify some things:

- Yes the lens has a large amount of barrel distortion.

- Video and photos (including RAW if using Lightroom/Photoshop) are corrected in camera or with an embedded lens profile so some do not ever see it.

- Raws if opended in some other editors (or stripped of a few exif tags for Lightroom) show the uncorrected original image off the sensor.

- There are two ways to correct the barrel distortion (a) with a lens profile (b) with a tool like lightrooms manual distortion correction:

(a) A lens profile does not correct, crop and up scale everything like my previous examples. It will stretch/scale outer edges (up scale) + push in the inner portion (down scale)

(b) LR lens tool manually will push it all in and then crop off warped edges. Still stretching edges but down scaling only overall.

DJI is most likely using method (a) in camera, not (b). Attached is a lens profile for Adobe LR or PS to do the same as (a) (but should only be used on Raws that have the embedded DJI profile stripped away first).

The lens (the weak spot vs the sensor) could certainly be improved upon for future models using a 1 inch sensor to enhance image quality further (and mainly around the edges which get quite soft and stretched).

Dingoz Posted at 2017-10-2 13:56
This might help clarify some things:

- Yes the lens has a large amount of barrel distortion.

yes, DJI RAWs contain a lens profile that RAW converters pick up automatically, so no need for seperate profile. however, what i was surprised about are 2 things:

1. P4pro takes full readout 19.9MP image off the sensor behind a ~20mm lens, then crops away as much as 15%, ending up with only ~17MP with FOV 24mm. Then they apply distortion correction which stretches out the corners & flattens image. finally they take those 17MP and scale them back up to 19.9MP.
all in all, it seems like a lot of back and forth for those pixels, especially toward the corners. Coming from photography world, something like that is a big red flag. i wonder if maybe skipping the scaling would have increased quality a bit. who needs those extra 3MP?

2. SONY RX100m4/m5 do the exact same thing! i compared uncorrected RAW, and it is only slightly less terrible than P4pro. RX100 is more or less undisputed best P&S of the last years, so apparently all this processing+scaling seems to be ok.

Sebb Posted at 2017-10-3 03:59
yes, DJI RAWs contain a lens profile that RAW converters pick up automatically, so no need for seperate profile. however, what i was surprised about are 2 things:

1. P4pro takes full readout 19.9MP image off the sensor behind a ~20mm lens, then crops away as much as 15%, ending up with only ~17MP with FOV 24mm. Then they apply distortion correction which stretches out the corners. then they take those 17MP and scale them back up to 19.9MP. all in all, it seems like a lot of back and forth for those pixels, especially toward the corners. Coming from photography world, something like that is a big red flag. i wonder if maybe skipping the scaling would have increased quality a bit. who needs those extra 3MP?

On point 1 I do not think that is the case... their lens profile just loses that much from the edges on its own without cropping and upscaling the whole lot. In fact the centre of the image is downscaled compared to an uncorrected raw. You can test this yourself by using that lens profile I attached on an uncorrected raw file in LR or PS. Toggling ti on/off and also comparing objects in the image to the unedited raw you quickly realise the outer is corrected up scaled and centre is actually downscaled.

That lens profile I attached does exactly the same thing and is almost identical to DJI's one in the amount it corrects, however it is a little less extreme while still removing the full barrel distortion which leaves you with a fraction more FOV (DJI's just goes a fraction further / a bit more stretching).

As for image quality alone and for web viewing at or below 13mp - it is arguably better to start with an uncorrected raw and use LR/PS alternative manual method (+100) which downscales overall and yes the image detail is sharper if the end result is downscaling again for web viewing.

4K on the P4p I've found to be pretty soft. 1080 is sharper, which doesn't make a lot of sense, but is true. This is the case for both h.264 and h.265. I've tried everything, and I always get better results shooting in 1080, which is fine with me most of the time. Very odd that the camera would be sharper at a lower data rate with less resolution, but that's the case. It's still a much better image than the previous generation of phantom.

fans80534a2e Posted at 2017-10-4 19:49
4K on the P4p I've found to be pretty soft. 1080 is sharper, which doesn't make a lot of sense, but is true. This is the case for both h.264 and h.265. I've tried everything, and I always get better results shooting in 1080, which is fine with me most of the time. Very odd that the camera would be sharper at a lower data rate with less resolution, but that's the case. It's still a much better image than the previous generation of phantom.

Unless you are viewing the image on a decent 4K display, you are at the mercy of the display scaling of the device you are viewing it on. I can look at the 4K output of my P3P on a very good 4K monitor and it is quite acceptable. I can look at the 4K output of my Sony A7rII camera and is is even better.

Hi, your problem is called Aliasing.

Check this out, skip to Sharpness section 08:25, but I recommend to watch the whole video:

mpaka Posted at 2017-10-5 04:20
Hi, your problem is called Aliasing.

Check this out, skip to Sharpness section 08:25, but I recommend to watch the whole video:

You're right, the effect of Aliasing is the same than Line Skipping, but technically are different, Aliasing is, how was described in the video, due to excess of sharpness that produce resolution reduction.

Why in my case is line skipping?... How you can see in the post #24 all the videos were recorded with exactly the same shooting and lighting conditions, the only technical difference is 4k@30 vs 4k@60, at 30fps is not visible the "aliasing" and at 60fps is completely visible, so is not aliasing due to extra sharpness, is line skipping due to some kind of technical issue that the P4P has when shoots at 60fps in 4K.

juvelezm Posted at 2017-10-7 07:01
You're right, the effect of Aliasing is the same than Line Skipping, but technically are different, Aliasing is, how was described in the video, due to excess of sharpness that produce resolution reduction.

Why in my case is line skipping?... How you can see in the post #24 all the videos were recorded with exactly the same shooting and lighting conditions, the only technical difference is 4k@30 vs 4k@60, at 30fps is not visible the "aliasing" and at 60fps is completely visible, so is not aliasing due to extra sharpness, is line skipping due to some kind of technical issue that the P4P has when shoots at 60fps in 4K.

"line skipping due to some kind of technical issue that the P4P has when shoots at 60fps in 4K."

Mate, you were lucky that the DJI Engineer did answer you in full detail.

the P4P is using RECTANGLE Pixels in 60p mode. This means that it is using logical Pixels - which are covering more height than width,  thus resulting in videos with for example 50% less usable vertical resolution.
You can also call this line skipping. But maybe they did choose Pixels with 1.5px logical height. so then lineskipping would be the wrong expression. Maybe they do use some sort of binning, too. Who knows.

Anyway, the technicians explained it nicely in two posts (#22 and #28), so everyone should notice:
4k@30p -> full resolution
4k@60p -> same fov, same noise performance, but reduced vertical resolution

This is acutal a behaviour seen in many, many cameras... higher framerate at same resolution often results in binning, line skipping or other ways of reducing the procedural load on the camera´s DSP.

Maybe its because you are using slow SD card?

Filip Kyzek Posted at 12-30 02:52
Maybe its because you are using slow SD card?

The SD card records digital information, therefore it cannot have any effect on the quality of the image recorded.

i am happy to have found the answer in this thread. post #22 did it all for me. bummer is that the 4k60 capability of the P4A was the reason i got it over the M2P