Mavic Pro-2 vs Zoom and P4P - Line Skipping + Sensor Heat?

[Dronefriend]

From blade-strike @ RCG:
"Engineering response to Binning / line skipping
Mavic 2 Pro uses a different processor than P4P and has two 4K modes, as stated on the product page, which are FOV mode and HQ mode. FOV mode has a wider view but less details than HQ mode. FOV mode's image processing requires raw subsampling, which scales 5.5K raw into 4K raw before feeding it to the image processing pipeline.
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.
DJI looked into many solutions for implementing FOV and found that the current FOV mode is the best compromise for the Mavic 2
hardware. Please remember the Mavic 2 is an extremely complex and compact piece of hardware. The current solution represents the best possible way to enable filming using the full 1” sensor wide angle area FOV on the Mavic 2 hardware."

Thanks for reposting. Do they have an explanation for the HQ mode as well?

 

[FiShBuRn]

@Dronefriend that's the full quote about the issue...

 

[Sockeye]

Awesome....can you post higher resolution clip of you test results?

Thanks!

Mavic 2 Pro Sensor Issues

 

[Dronefriend]

They should have done a 1 inch sensor with 1080p resolution. Very big pixels with good light intake. And a 1:1 read-out.

 

[chunky]

What should „RAW subsampling“ be ;-)?!
This answear is a joke!
If the complete 5.5K sensor is read we would have nothing else than with the P4Pro that downsamples the 5.5K image to 4K output.
As many tests state the 4K FFOV mode only gives resolution of close to 2.7K real image information, horizontal as vertical.
With simple maths this is half the 5.5K resolution of the sensor, so the sensor resolution is divided by two - maybe by lineskipping or pixelbinning (I expect the later since SNR seems to improve a bit in FFOV).

We have some people here that are somewhat deeper in the techs, don‘t expect everyone to be plain stupid.

 

[mrg]

What should „RAW subsampling“ be ;-)?!

It is given in the response.

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 ... optimized ... more advanced than the simple average mechanism of pixel binning.

In short: "Raw subsampling" is not simple average pixel binning, it's a optimized and more advanced pixel binning!

 

[chunky]

Specsheet to the possibly used sensor:
https://www.sony-semicon.co.jp/products_en/IS/sensor0/img/product/cmos/IMX183CLK-J_CQJ-J_Flyer.pdf

What we learn is that the sensor can be windowed/cropped and so gives us in 16:9 the following framerates (according Table 3):

16:9 height: 5472 / 16 x 9 [px] = 3078px,
so 16:9 image is 5472 x 3078 [px],
what is 3648 / 3078 [px] = 1.185 times faster for readout than the full 3:2 sensor area.

So we would get 24.98 x 1.185 = 29.6 [fps] in 10bit and 21.98 x 1.185 = 26 [fps] in 12bit - at least enough for creating „real 10bit from 12bit“ in 24p or 25p!

We don‘t know what‘s going on in the P4pro that it‘s possible to get full scan 30p that would only be a tiny bit higher than the „specified“ 29.6p. Maybe the Input/LVDS frequency is a bit higher from the (SoC’s) PLL, „maybe the older sensor“ in the P4p is a bit different.

BTW: nearly everyone found the FFOV resolution to be lower than that from the P4p, why should this only been seen in testcharts (which not only represent a single color at least in one case), not in real recordings?!
At least „Cliff...“ and also me compared „real recordings“ and we both come to the same conclusion, the M2P lacks behind!
You see this in grass and trees/branches and also in diagonal lines which show soft since interpolated humps!!

DJI, don‘t think us stupid!
Deliver what you promised us in the manual and the FAQs - (close) „4K resolution“, not only imagesize!

 

[chunky]

It is given in the response.

In short: "Raw subsampling" is not simple average pixel binning, it's a optimized and more advanced pixel binning!

So why should „this“ be used when it was not necessary with the P4pro and - binning and „advanced“ BINNING - what‘s behind this? Who cares?
We loose resolution that is promised in the manual and the FAQ‘s - we have NO (not closely) 4K resolution, we only get nearly 2.7K so this „advanced binning“ shows no benefit (in resolution)!

When DJO talks about „binning“ we only get interpolated 2.7K to 4K image what is no 4K resolved image and also has no groundwork for 4K! I didn‘t buy a 2.7K camera!!
I was offered a 4K camera and when DJI talks about „binning“ it‘s clear that there is no 4K resolution any more present!
There is no „magic“ „advanced binning“ known for bayer sensors that retains the native (5.5K) resolution, DJI also wins no nobel price here but trouble since nobody sees a benefit, not in testcharts nor in real life captures!

 

[spence]

I have to agree, Chunky. No one is asking for anything unreasonable, here, just video and stills that are comparable to the P4P, which uses the same size sensor.

This reeks of intentional crippling for product differentiation, to me. I bet you'll see a new drone from DJI in a year or 18 months which somehow, amazingly, manages to provide P4P level quality from that 1" sensor. At a slightly upgraded price, of course.

 

[mrg]

So why should „this“ be used when it was not necessary with the P4pro and - binning and „advanced“ BINNING - what‘s behind this? Who cares?
We loose resolution that is promised in the manual and the FAQ‘s - we have NO (not closely) 4K resolution, we only get nearly 2.7K so this „advanced binning“ shows no benefit (in resolution)!

The simple pixel binning of the Mavic 1 Pro by more than 30 fps had a very bad quality, which caused strong color flickering on all contrast edges. The new advanced pixel binning is much better but also softer.

 

[Sockeye]

a M2P Platinum?

 

[mrg]

a M2P Platinum?

Or a Mavic 2 Enterprise they want to sell too.

 

[chunky]

I want what they promoted - (close to) 4K „resolution“, no hyper magic advanced binning blablabla that‘s nothing but interpolated 2.7K.
Especially in wideangle resolution is necessary, I can live with the cropped „only native“ HQ mode for the longer 42mm.

 

[yysc]

Can the sensor be actually read-out with 8Bit? Not sure...

The sensor can read 12,10,9,8, etc. Below MediaInfo in one of my M2P clips in 4K, when you export in H264 the video is 8 bit.

Complete name : E:\m2p\DJI_0008.MP4
Format : MPEG-4
Format profile : Base Media
Codec ID : isom (isom/iso2/avc1/mp41)
File size : 649 MiB
Duration : 52 s 686 ms
Overall bit rate : 103 Mb/s
Encoded date : UTC 2018-09-02 13:01:48
Tagged date : UTC 2018-09-02 13:01:48
Writing application : Lavf56.15.102

Video
ID : 1
Format : AVC
Format/Info : Advanced Video Codec
Format profile : [email protected]
Format settings : CABAC / 1 Ref Frames
Format settings, CABAC : Yes
Format settings, RefFrames : 1 frame
Format settings, GOP : M=1, N=29
Codec ID : avc1
Codec ID/Info : Advanced Video Coding
Duration : 52 s 686 ms
Bit rate : 103 Mb/s
Width : 3 840 pixels
Height : 2 160 pixels
Display aspect ratio : 16:9
Frame rate mode : Constant
Frame rate : 29.970 (30000/1001) FPS
Original frame rate : 29.000 FPS
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
Bits/(Pixel*Frame) : 0.416
Stream size : 649 MiB (100%)
Encoded date : UTC 2018-09-02 13:01:48
Tagged date : UTC 2018-09-02 13:01:48

 

[yysc]

From blade-strike @ RCG:
"Engineering response to Binning / line skipping
Mavic 2 Pro uses a different processor than P4P and has two 4K modes, as stated on the product page, which are FOV mode and HQ mode. FOV mode has a wider view but less details than HQ mode. FOV mode's image processing requires raw subsampling, which scales 5.5K raw into 4K raw before feeding it to the image processing pipeline.
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.
DJI looked into many solutions for implementing FOV and found that the current FOV mode is the best compromise for the Mavic 2
hardware. Please remember the Mavic 2 is an extremely complex and compact piece of hardware. The current solution represents the best possible way to enable filming using the full 1” sensor wide angle area FOV on the Mavic 2 hardware."

I call the highlighted in red from DJI's reponse BS, go to 5:50 in the video below, the M2P has much higher aliasing and less definition than the Autel Evo. And these are not charts.

The M2P in 4K FOV has the clarity of the Mavic Air at double the cost. Well yes, DR and noise is better, I guess this is the consolation prize.

 

[chunky]

The sensor can read 12,10,9,8, etc. Below MediaInfo in one of my M2P clips in 4K, when you export in H264 the video is 8 bit.

Complete name : E:\m2p\DJI_0008.MP4
Format : MPEG-4
Format profile : Base Media
Codec ID : isom (isom/iso2/avc1/mp41)
File size : 649 MiB
Duration : 52 s 686 ms
Overall bit rate : 103 Mb/s
Encoded date : UTC 2018-09-02 13:01:48
Tagged date : UTC 2018-09-02 13:01:48
Writing application : Lavf56.15.102

Video
ID : 1
Format : AVC
Format/Info : Advanced Video Codec
Format profile : [email protected]
Format settings : CABAC / 1 Ref Frames
Format settings, CABAC : Yes
Format settings, RefFrames : 1 frame
Format settings, GOP : M=1, N=29
Codec ID : avc1
Codec ID/Info : Advanced Video Coding
Duration : 52 s 686 ms
Bit rate : 103 Mb/s
Width : 3 840 pixels
Height : 2 160 pixels
Display aspect ratio : 16:9
Frame rate mode : Constant
Frame rate : 29.970 (30000/1001) FPS
Original frame rate : 29.000 FPS
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
Bits/(Pixel*Frame) : 0.416
Stream size : 649 MiB (100%)
Encoded date : UTC 2018-09-02 13:01:48
Tagged date : UTC 2018-09-02 13:01:48

Uhh, the encoders (SoC) output stream has nothing to do with the sensor readout bitdepth.
You can also make an 8bit MPEG2 from a 16bit REDRAW clip.

But, even for 8bit output the sensor should (must) be read in at least 10bit to perform „lossless“ gamma adaption (D Log M/HLG), so for real 10bit output we should expect 12bit sensor readout. Otherwise we would get less than real 10bit graduation.

 

[yysc]

Uhh, the encoders (SoC) output stream has nothing to do with the sensor readout bitdepth.
You can also make an 8bit MPEG2 from a 16bit REDRAW clip.

But, even for 8bit output the sensor should (must) be read in at least 10bit to perform „lossless“ gamma adaption (D Log M/HLG), so for real 10bit output we should expect 12bit sensor readout. Otherwise we would get less than real 10bit graduation.

Ok, thanks, but I'm still not convinced on why the M2P cannot render video just like the P4P, when stills rendering is faily similar. I now feel they are arbitrarily making the M2P worse so they can protect the P4P 2.0.

A shame and misleading for the customers. Hasselblad means almost nothing here.

 

[chunky]

Hardly anyone of us but DJI knows why FFOV video is „designed“ to be so much worse than P4pro...

 

[yysc]

2. the sensor overheats - also hardly believeable since the P4pro‘s camera isn’t bigger and wouldn‘t dissipate heat better.
The P4pro‘s camera may have better cooling due to better air stream from the props, maybe, than limitate full sensor readout by temperature as Sony does with the RX100V and monitor this in the app.
Before temperature gets to high simply make a fall back to binning. Sony‘s RX100V does up to 5mins. in full sensor readout without (!!) active cooling - not to bad for many things and not to „professional“ - I „could“ live with that.

The 5 min limit is SONY's own design issue, Panasonic LX10 (similar in size to RX100) also uses SONY 1" BSI 20MP sensor and such limitation does not exist in 4K.

Also GH5 using SONY sensor does not overheat and many of the competitor SONY cameras do.

 

[chunky]

The 5 min limit is SONY's own design issue, Panasonic LX10 (similar in size to RX100) also uses SONY 1" BSI 20MP sensor and such limitation does not exist in 4K.

Also GH5 using SONY sensor does not overheat and many of the competitor SONY cameras do.

Ah, that‘s not the same: LX10 has lower 4K resolution than RX100V since it‘s also only „native“ cropping the sensor in 4K so you get 36mm instead of 24mm for photo.
RX100V uses the full width with oversampling like the P4pro.

Image comparison: Digital Photography Review

GH5 is a complete different animal with a M43 sensor and a big body that can dissipate temperature better.
Question is what happens with heat in the P4pro in 4K oversampling video when not cooled by the props when filming on the ground.
I have mine not here (I’m on vacation) to test.