They call it "Raw subsampling ", whatever that is. But they surely need to optimize the algorithm...

Yeah.. Subsampling, undersampling whatever it's binning with digital LPF to limit aliasing and limit the noise.
It's better than line skipping, but the result is quite the same in terms of resolution
The P4P makes oversampling and has a 4K resolution better than a 12MP
The MP2 in 'cropped HQ' is a 12MP so has the same resolution
The MP2 in 'Full FOV' is a binned one with a resol lower than a 12 MP (eq. 2.7 K)
I like the answer of the guy on RCG : 'it's visible only on test chards at 400% , not on real images (I guess, on the sky)' .

Downscale the image by 4x and you will still see it... 400% just makes it also visible for iPhone users... ;-)

castormalin Posted at 2018-9-18 09:22
Yeah.. Subsampling, undersampling whatever it's binning with digital LPF to limit aliasing and limit the noise.
It's better than line skipping, but the result is quite the same in terms of resolution
The P4P makes oversampling and has a 4K resolution better than a 12MP

Actually the mp2 could still do 8.3MP (pixels needed for 4K), if the P4P makes oversampling, it will be better.

I think the resol of the MP2 is closer to what you will get with a 6MP ... Not better than basic binning (20MP/4). The only good point is that it includes a digital LPF to reduce color aliasing.
But it's quite annoying to get a 20MP for a lower resolution than the small 12MP (in the FOV mode)
Like buy a V8 to get finally a car that works on 4 legs

Ok guys, it‘s simple. DJI claims in the manual on p.38 and in the FAQs for the FullFOV mode that the 5.5K sensor is down sampled to a 4K „resolution“ (!) image.
Nobody and also not DJI can currently show that there is nearly 4K „resolution“ - there is only 4K imagesize what is a complete different story!
What should this „magic“ advanced binning be?! Does it resolve as the P4pro did?! No, you will never get close to 4K image resolution that the P4p won through oversampling from a „binned“ ~2.7K source!
We simply lose visible detail!

Yep, they can call it whatever they want, but the proof is in the pudding. And that pudding shows an image barely better than the Mavic Air, and well behind the P4P. Very disappointing.

i would like an official statement from Dji about this.

I guess if it's not a question of HW architecture, DJI could at least propose what the P4P does in 8bit format, then the people would have the choice: dynamics vs (resolution + lens correction) in Full FOV

CineView Media Posted at 2018-9-18 12:50
i would like an official statement from Dji about this.

I just realize (if I'm not wrong)  that the answer of Blade Strike / DJI Entreprise Support on RCG has not been posted here:
https://www.rcgroups.com/forums/ ... &postcount=2615

castormalin Posted at 2018-9-18 13:17
I just realize (if I'm not wrong)  that the answer of Blade Strike / DJI Entreprise Support on RCG has not been posted here:
https://www.rcgroups.com/forums/showpost.php?p=40238048&postcount=2615

Interesting but contradicting: if the FOV mode does anything better than line skipping, it should deliver better results than the HQ mode by definition (=more pixels, more information), which it doesn't.

castormalin Posted at 2018-9-18 13:17
I just realize (if I'm not wrong)  that the answer of Blade Strike / DJI Entreprise Support on RCG has not been posted here:
https://www.rcgroups.com/forums/showpost.php?p=40238048&postcount=2615

Quoting that:

"Raw subsampling is different from line skipping; it uses the full area (of the 16:9 portion) of the image sensor so the signal-noise-ratio is much better than line skipping. Raw subsampling is also different from pixel binning. It is an optimized filtering & sampling process which is much more advanced than the simple average mechanism of pixel binning. Raw subsampling causes some loss in the finest details. However, this is not as visible in real scenes (like aerial filming) as it is on charts. "



In this comparison video, at marker 5min:50secs, one can see very obvious aliasing in the building's roof in the M2P (recorded in 4K FOV) whereas it is much minor in the Evo's. I may have Bionic eyes and mere mortals don't see this or maybe DJI engineers need to get their eyes checked.

...as to be seen here before

Roland Schulz Posted at 2018-9-18 15:05
...as to be seen here before

Thanks! My p4p is definitely a keeper then!

Btw, aliasing patterns might well be absent from a real-life image, side effects of the line-skipping process cannot be avoided: less details, dynamic range and more noise due to the smaller amount of data.

As for the mp2, it needs to find its place on the market. I think it is still very much superior to the mp! If someone needs a really portable solution and has no space for the Phantom, this is still the way to go.
If I have the energy, time and battery charge to waste I'll try to add a comparison of my P4P with a sony rx100iv, one of the first 1 inch cameras doing real full-sensor readout.

I think using Hasselblad 1inch sensor as selling point  just Marketing trick! Because you not get result of 1inch sensor in the end!
Everyone can tell that you will not see result in social media or youtube etc. But why than there exists Inspire etc.?
Yes, it is portable with great camera for that size, but it is not with quality of 1inch sensor!

Andrejs Posted at 2018-9-18 21:58
I think using Hasselblad 1inch sensor as selling point  just Marketing trick! Because you not get result of 1inch sensor in the end!
Everyone can tell that you will not see result in social media or youtube etc. But why than there exists Inspire etc.?
Yes, it is portable with great camera for that size, but it is not with quality of 1inch sensor!

There are limiting factors, that we have to be aware of, before jumping on any marketing activity/hype, mostly basic physics:

There is the a lens limitation: to illuminate a 1" sensor properly at a decent EV (should be around f2.0) with less waste on the borders you need a fair amount of glass, probably 4x the weight of the current optics.

There is the processing power limitation: full-sensor readout at 4k60p on an ~18MP area and subsequent downsizing requires a LOT of processing power. Currently there are no compact cameras on the market which can do that! It's the level of the Panasonic GH5! Full-sensor 4k30p is still challenging, cameras doing that (rx100iv-vi, panasonic LX10/100) produce a lot of heat and eat their batteries fast.
Being realistic, an upgrade from the current levels would be an inspire with x5s or x7.

Nobody talks about 4K60p here, we talk about that what was possible with the P4pro and what‘s specified by DJI as 4K „resolution“, not imagesize.

Andrejs Posted at 2018-9-18 21:58
I think using Hasselblad 1inch sensor as selling point  just Marketing trick! Because you not get result of 1inch sensor in the end!
Everyone can tell that you will not see result in social media or youtube etc. But why than there exists Inspire etc.?
Yes, it is portable with great camera for that size, but it is not with quality of 1inch sensor!

Its not a Hasselblad sensor, its a Sony.
The only Hasselblad component in this camera, is the color science, rest is Dji manufactured as the previous ones.

So, if 4k on this camera is the same as 2.7k, then we might as well save some recording space and record "4k" in 2,7k 60p instead then, i mean, we still get "4k" 60p that way.

CineView Media Posted at 2018-9-19 03:00
Its not a Hasselblad sensor, its a Sony.
The only Hasselblad component in this camera, is the color science, rest is Dji manufactured as the previous ones.

Yeah you are right, I am was meaning Hasselblad camera with 1inch sensor Ok, with name on it.

CineView Media Posted at 2018-9-19 03:02
So, if 4k on this camera is the same as 2.7k, then we might as well save some recording space and record "4k" in 2,7k 60p instead then, i mean, we still get "4k" 60p that way.

Yes, good point. No reason to shoot 4k than if we can make upscale later to 4k and get same result but with 60fps ))

CineView Media Posted at 2018-9-19 03:02
So, if 4k on this camera is the same as 2.7k, then we might as well save some recording space and record "4k" in 2,7k 60p instead then, i mean, we still get "4k" 60p that way.

Havent verified myself, but others state 2.7K60p is even worse than 2.7K30p!

I rather have some rolling shutter and a much sharper image then a softer image with no rolling shutter. The drone is not usually filming fast action sports anyways, but slow moving landscapes instead.

Ruben11 Posted at 2018-9-19 07:51
I rather have some rolling shutter and a much sharper image then a softer image with no rolling shutter. The drone is not usually filming fast action sports anyways, but slow moving landscapes instead.

Sorry but not everyone filming only landscapes I use it for some extreme sport filming too

Andrejs Posted at 2018-9-19 10:25
Sorry but not everyone filming only landscapes  I use it for some extreme sport filming too

I guess you don't shoot fast action at 4k30p...

I find it very hard to believe that they supposedly spent 2 years developing this camera, and then just before the release, they discovered a heat problem. If they are  protecting Phantom 2 sales, well , now with the recent price increase, the Phantom is cheaper, so maybe they will "unlock" the camera to full resolution.

St-even Posted at 2018-9-20 00:03
I find it very hard to believe that they supposedly spent 2 years developing this camera, and then just before the release, they discovered a heat problem. If they are  protecting Phantom 2 sales, well , now with the recent price increase, the Phantom is cheaper, so maybe they will "unlock" the camera to full resolution.

There won't be any unlock, the heat production and power demand of full-sensor readout hasn't changed much in the last 2 years, the mp2 is basically on the same sensor and likely a very similar SoC.

One thing, maybe the 10bit readout causes the problem (10bit means substantially more data to read). Does the mp2 have real 8bit mode, when only 8bits are read from the CMOS and the "unnecessary" 2 bits are not omitted later in the pipeline? Has it been compared to the crowd?

FlyingUser Posted at 2018-9-20 02:17
There won't be any unlock, the heat production and power demand of full-sensor readout hasn't changed much in the last 2 years, the mp2 is basically on the same sensor and likely a very similar SoC.

One thing, maybe the 10bit readout causes the problem (10bit means substantially more data to read). Does the mp2 have real 8bit mode, when only 8bits are read from the CMOS and the "unnecessary" 2 bits are not omitted later in the pipeline? Has it been compared to the crowd?

This is the SONY IMX183 datasheet
https://www.sony-semicon.co.jp/p ... _2014/imx183_e.html

It says it can read the entire 3:2 frame (20mp) at 10bit 25times per second. The 16:9 crop of 20mbps is about 16.6mp. It may be possible for that sensor to do 16.6mp@10bit at 30fps.

Unless the SoC cannot deal with this data (the P4P with the older Amba H1 can) I can't explain why there are doing this terrible subsampling.

And then 1080p@60fps is "moire-land". It does worse than the Mavic 1 Pro after the moire issue was fixed in drones produced from Summer 2017 onwards

yysc Posted at 2018-9-20 11:42
This is the SONY IMX183 datasheet
https://www.sony-semicon.co.jp/products_en/new_pro/may_2014/imx183_e.html

Maybe they saved a few grams of copper in the cooler?  Why are the stills noisier? Hotter sensor??

as far as I remember,  the dark noise is x2 every ~ 6°C.

castormalin Posted at 2018-9-20 13:34
as far as I remember,  the dark noise is x2 every ~ 6°C.

That might be the case. Smaller bird, tighter, hotter package:

It's not all what it seems, the p4p might be adding a little noise filtering.

Well, this is a bit of a shame if its true... but after reading the DJI explanation of how the M2P 1" sensor subsampling works, its possible that the firmware could be updated with a better subsampling process if and when one is developed. As of right now, the current process isn't very good.

I still love the drone, but I'm pretty disappointed that its not the same quality as the P4P, which is what I believe everyone was expecting when DJI announced this bird.

Roland Schulz Posted at 2018-9-19 04:48
Havent verified myself, but others state 2.7K60p is even worse than 2.7K30p!

Yes, 2.7K 60p is much worse than 2.7K 30p!

Exactly... I don’t know why they couldn’t of just put the guts out of the 4 pro into a Mavic platform and called it a day . Right now we are stuck with a Mavic 2 Pro with inferior camera performance than the 4 pro at a higher price.

El Diablo Posted at 2018-9-21 00:25
The same thing happens in photography all the time... people spend hours shooting wall charts and brick walls instead of real life situations... That test only proves that the M4P is sharper at a couple of feet away, The only lenses designed for shooting flat test charts are macro! Don't believe me? Get a 4.000$ Zeiss 55mm Otus and shoot a chart against a 280$ Nikkor 40mm Micro, I bet the Nikon will destroy the Zeiss.

The question you need to ask is?

WetalkUAV tested under real conditions on a beach.

El Diablo Posted at 2018-9-21 00:51
And you were able to see that difference unless zoomed at 400%...?

For me the M2P's 4K is better than the Mavic Pro / Air and a slightly less sharp than the P4P, adding the difference in portability made my decision quite easy!

Nope, checking 100%. P4P still more crisp.

So I'm just better buying the zoom? I have both on cart right now and I can't decide. HELP!