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

[SteveCA]

Now we just need someone with a ground mounted, 200m by 350m test chart...

I can find the chart if someone can provide a P4Pro V2.0. I have MP 2 Pro ND’s to test with as well.

 

[chunky]

Here we see DJI still cheats us with wrong specs on their website. We are still told we get 4K resolution from the camera - they still mix up with „image size“ or „video mode“ and they are to arrogant even to change this here!

 

[WithTheBirds]

Here we see DJI still cheats us with wrong specs on their website. We are still told we get 4K resolution from the camera - they still mix up with „image size“ or „video mode“ and they are to arrogant even to change this here!
View attachment 48196
View attachment 48197

Why wouldn’t downsampling to 4K resolution qualify as 4K? Seems fair....

 

[MrSpaceman]

Here we see DJI still cheats us with wrong specs on their website. We are still told we get 4K resolution from the camera - they still mix up with „image size“ or „video mode“ and they are to arrogant even to change this here!
View attachment 48196
View attachment 48197

How would you define 4K?

 

[chunky]

Why wouldn’t downsampling to 4K resolution qualify as 4K? Seems fair....

If we would get 4K „resolution“ instead of only imagesize this would be fair - of course!
You know the difference between resolution and imagesize-/dimension?!

In „true“ 4K (3840x2160) every single pixel can adopt to an individual state not influenced by surrounding pixels.
An ideal 4K camera could represent up to 1920 linepairs or 3840 lines horizontal at the position defined by the pixel output matrix (...where the sampling happens. If the input is not aligned Nyquist theoreme kicks in and you only get half).

Since we mostly have bayer cameras, non perfect lenses and so on this is hardly achievable. Oversampling at least brings us somewhat closer to 4K image detail as to be seen with the P4pro.
This is the reason very most companys state their products with 4K imagesize, 4K modes and so on.
DJI states „resolution“ and the resulting <2.7K is far from 4K.

We already had this before...

 

[yysc]

Yes FOV mode is 4K video in size but not in resolution.

 

[sar104]

If we would get 4K „resolution“ instead of only imagesize this would be fair - of course!
You know the difference between resolution and imagesize-/dimension?!

In „true“ 4K (3840x2160) every single pixel can adopt to an individual state not influenced by surrounding pixels.
An ideal 4K camera could represent up to 1920 linepairs or 3840 lines horizontal at the position defined by the pixel output matrix (...where the sampling happens. If the input is not aligned Nyquist theoreme kicks in and you only get half).

Since we mostly have bayer cameras, non perfect lenses and so on this is hardly achievable. Oversampling at least brings us somewhat closer to 4K image detail as to be seen with the P4pro.
This is the reason very most companys state their products with 4K imagesize, 4K modes and so on.
DJI states „resolution“ and the resulting <2.7K is far from 4K.

We already had this before...

No - you really don't want each individual pixel to be independent - that would lead to wicked aliasing. The issue with FOV mode, if there is one, is that DJI has taken a different approach to the spatial filtering that always has to be applied to prevent it.

 

[sar104]

Yes FOV mode is 4K video in size but not in resolution.

There is no such thing as "4k resolution".

 

[Cliff_622]

I couldn't resist doing a few, in my opinion, more controlled chart tests on these issues.

Using a test chart designed to detect aliasing, and with the chart the same distance from the camera for all modes, since the optics don't change and therefor the size of the image on the sensor doesn't change with mode, I acquired DNG images, HQ video in Dlog-M, and FOV video in Dlog-M. For reference I shot the same image with a Sony RX100IV with the zoom set to the same as the M2P. All images were ISO 100 and F4.

The following files are cropped, resampled as explained below, but not processed in any other way, except that the final cropped versions are converted to jpeg (indistinguishable from the DNGs and 16-bit tiff captures from the videos) from a resolution point of view.

Comparison of the M2P DNG with the RX100 DNG shows similar resolution - not unexpected in terms of sensor properties since they presumably have similar sensors, but it means that the M2P lens is not too shabby either.

View attachment 48181
M2P DNG

View attachment 48182
RX100 DNG

The slightly angled line pairs are the best indicator of aliasing, and both these images start to show aliasing at about the same line density - level 6. As a sanity check, combining channels and running a line profile on the horizontal line pairs for the DNG produces the following:

View attachment 48185

This clearly shows the loss of contrast at around level 6.

HQ video is a simple crop of the central 4k pixels of the chip, read out fully. It displays similar resolution to the full-chip DNG - expected since it is not sub-sampling - it's fully sampling a portion (3840 x 2160) of the full chip.

View attachment 48183
HQ video, Dlog-M

FOV video is some kind of sub-sampling, and shows some very strange artifacts:

View attachment 48184
FOV video, Dlog-M

Note, in particular, the fact that level 4 shows much more aliasing than levels 5, 6 or 7. That levels 6 and 7 are resolved at all suggests to me that it is not pixel-binning or sub-sampling the sensor sites, but that some kind of spatial filtering is happening on the sensor chip before readout - sort of consistent with some of the comments coming out of DJI on the subject.

For comparison I took the full-sensor DNG and downsampled it to 4k resolution:

View attachment 48186
DNG downsampled to 4k

The result is interesting, but not unexpected relative to the HQ video/DNG. Level 4 is much better relative to the FOV video, while levels 5, 6 and 7 all show more aliasing. So in some ways, DJI's processing of the FOV video brings out more detail, but at the expense of some strange artifacts at intermediate spatial frequencies.

I also shot some real-world images and video and it seems to me that the problems seen in the resolution chart results are really difficult to discern in real video. The DNG still images are competitive with the RX100, which I would count as a significant success.

CAVEATS:

This only presents data from near the center of the field of view, and at ISO 100, F4. Subjectively there was not a significant problem with edge softness, but I haven't done resolution chart tests at the edges. Noise seems well controlled at ISO 100, but I did not test higher-gain settings.

Question: Did you say that you did NOT adjust your camera distance to the wall for each of these camera focal lengths? If so, why would NOT re calibrate each lens distance to the chart as required by the chart? The chart calibration corners need to be honored for any lens that is tested on it. Every lenses magnification factor needs to be canceled out. Generally this is done by calibrating it's distance to the chart and re calibrating it for another focal length. Instead, you have maintained the the same wall distance and and "equalized" the focal length by cropping the optically magnified captured image result?

If so,...that's a bit of a "reverse" way of doing it. The industry standard way would be to align charts optically then capture instead of cropping in and aligning it as a post or "after the capture" technique.

I suppose that cropping afterwards could in theory achieve the same result but it's generally more common to align optically (distance adjustment) instead.

CT

 

[Cliff_622]

There is no such thing as "4k resolution".

What do you mean?

"4k" certainly has a maximum mathematical resolution which is it's ultimate Nyquist frequency limit.

As a "container" it has a maximum amount of water it can cold. Weather you have it 1/2 filled or not is another question.

CT

 

[mcsenerd]

...I like debating and I honestly feel like this isn't DJI's fault. This is on YouTube. ...

This is where we "sorta" diverge. Exactly who sends these YT cats these things on a regular basis? You can't possibly completely absolve DJI, when these guys you're talking about are almost essentially THE core part of their marketing strategy and one that they've cultivated for years. So think about it for a moment...and this is an honest question here, does DJI continue to send pre-release drones to folks who are critical of them or do they "thin the herd" and keep the most fanatical? One of the magical things behind this marketing strategy is the wonderful deniability it produces, because you get to reap the rewards of the hype with none of the direct liabilities when it goes too far and the words don't come out of you directly...

 

[sar104]

Question: Did you say that you did NOT adjust your camera distance to the wall for each of these camera focal lengths? If so, why would NOT re calibrate each lens distance to the chart as required by the chart? The chart calibration corners need to be honored for any lens that is tested on it. Every lenses magnification factor needs to be canceled out. Generally this is done by calibrating it's distance to the chart and re calibrating it for another focal length. Instead, you have maintained the the same wall distance and and "equalized" the focal length by cropping the optically magnified captured image result?

If so,...that's a bit of a "reverse" way of doing it. The industry standard way would be to align charts optically then capture instead of cropping in and aligning it as a post or "after the capture" technique.

I suppose that cropping afterwards could in theory achieve the same result but it's generally more common to align optically (distance adjustment) instead.

CT

You need to move the camera distance if you are changing the lens characteristics. The lens focal length is constant on the M2P, so changing the distance for different crops is neither valid nor useful. The chart corners need to be in the same place relative to the image on the sensor. As in your tests - this is not using the chart to fill the field of view of the lens/sensor combination - it is to get relative resolution values.

 

[Lon Denard]

Now we just need someone with a ground mounted, 200m by 350m test chart...

Hahaha!

 

[sar104]

What do you mean?

"4k" certainly has a maximum mathematical resolution which is it's ultimate Nyquist frequency limit.

As a "container" it has a maximum amount of water it can cold. Weather you have it 1/2 filled or not is another question.

CT

No - that is a meaningless analogy. There is nothing in any 4k video standard about resolution. 4k refers to image dimension. Now stating that a 4k image has a maximum resolution determined by the Nyquist limit is correct, but not part of any standard or even useful, since an image achieving that would have inadequate spatial filtering.

 

[Lon Denard]

You won't get a lot of comments here SAR- you haven't included assumption and speculation..... '-)

Thank you for doing this- very interesting indeed.

It would seem DJI has been honest and there is probably no conspiracy to deceive or otherwise cheat their customers.

WHAT?!? DJI FAN BOY ALERT! How much is DJI paying you? Hahaha! That's exactly what I've been saying. The camera may be a disappointment to some, that's subjective, but the rest of this evil motive based supposition is ridiculous.

 

[sar104]

You won't get a lot of comments here SAR- you haven't included assumption and speculation..... '-)

Thank you for doing this- very interesting indeed.

It would seem DJI has been honest and there is probably no conspiracy to deceive or otherwise cheat their customers.

It's unfortunate - in many ways this is a very useful and valid discussion but it is quite shocking how little many people understand about the subject, and where they seem to get their information. As a result, most of the posts are just assertions, counter-assertions and speculations fueled by random YouTube videos. I've no idea in what way having a "YouTube presence" or whatever it is called gives any specific credibility to the content. No qualifications needed, scientific method deprecated, and no peer review required, but plentiful fodder for confirmation bias - if you want to argue a position you will almost certainly find a YouTube channel to support it. A simple search on Wikipedia would be enough to dispel many of these notions, but I guess that leaves no room for all those "beliefs".

 

[Lon Denard]

This is where we "sorta" diverge. Exactly who sends these YT cats these things on a regular basis? You can't possibly completely absolve DJI, when these guys you're talking about are almost essentially THE core part of their marketing strategy and one that they've cultivated for years. So think about it for a moment...and this is an honest question here, does DJI continue to send pre-release drones to folks who are critical of them or do they "thin the herd" and keep the most fanatical? One of the magical things behind this marketing strategy is the wonderful deniability it produces, because you get to reap the rewards of the hype with none of the direct liabilities when it goes too far and the words don't come out of you directly...

Perhaps I was too non specific. All the pre hype, if you will. The going over every leaked photo. That Ostalav(sp) garbage. Every wild and idiotic video speculating about what the drone will do.

As I'm sure you know, everyone with pre production unit were under NDA's. They weren't part of the speculation frenzy that preceded this drone's launch. Heck, they've already moved on to the P5P! DJI didn't say a word up until launch day.

Now certainly after launch, there was a lot of gushing going on but they weren't pixel peeping or pounding the specs, for the most part. However, there were a few early videos that talked about specs, like MadRC's that I've mentioned several times. I'm sorry if Casey or iJustine made your buying decision for you.... I read the manual, I checked out the specs, I watched the more technical channels like MadRC and Tom's Tech Time and I purchased the drone.

One of the funniest impressions I've gotten from this thread is that some of you apparently expect DJI and presumably every company, to list any and all faults or "missing features (aka features YouTube speculators told you it was going to have) of their product during their launch events or on their website.

I'm a car and motorcycle guy, so I use car and motorcycle analogies. Sorry, here's another one...

My GSXR1000 does about 180 mph. You would demand that Suzuki explain why it doesn't do 200, point out that it didn't get the posted 30MPG at that speed, the wind noise levels and be accountable for Sport Riders' article claiming it would do 0-60 in 3 seconds instead of the actual 3.3 seconds that it takes. I think that's unreasonable or at the very least, unrealistic.

 

[Cliff_622]

Perhaps I was too non specific. All the pre hype, if you will. The going over every leaked photo. That Ostalav(sp) garbage. Every wild and idiotic video speculating about what the drone will do.

As I'm sure you know, everyone with pre production unit were under NDA's. They weren't part of the speculation frenzy that preceded this drone's launch. Heck, they've already moved on to the P5P! DJI didn't say a word up until launch day.

Now certainly after launch, there was a lot of gushing going on but they weren't pixel peeping or pounding the specs, for the most part. However, there were a few early videos that talked about specs, like MadRC's that I've mentioned several times. I'm sorry if Casey or iJustine made your buying decision for you.... I read the manual, I checked out the specs, I watched the more technical channels like MadRC and Tom's Tech Time and I purchased the drone.

One of the funniest impressions I've gotten from this thread is that some of you apparently expect DJI and presumably every company, to list any and all faults or "missing features (aka features YouTube speculators told you it was going to have) of their product during their launch events or on their website.

I'm a car and motorcycle guy, so I use car and motorcycle analogies. Sorry, here's another one...

My GSXR1000 does about 180 mph. You would demand that Suzuki explain why it doesn't do 200, point out that it didn't get the posted 30MPG at that speed, the wind noise levels and be accountable for Sport Riders' article claiming it would do 0-60 in 3 seconds instead of the actual 3.3 seconds that it takes. I think that's unreasonable or at the very least, unrealistic.

Thats a great post and I have to agree with much about what you say. I believe are partially to blame about the hype that was created over the Mavic-2 Pro. For almost a year WE wanted the M2P to be the P4P killer. And,..when we found out that it was going to have a 1 inch-type sensor, we automatically went ape-$hit happy that DJI finally did this for us! Wahoo...down the P4P! We got the P4P in a Mavic now! Yeah!!!!!

But DJI was NEVER going to burn a P4P V2.0 that JUST hit the market a few months ago!! That was clearly never their goal.

So now, many if us feel stupid and silly for ever thinking DJI was going to do this. And,..we blame DJI for our overly-high expectations being dashed.

DJI is out to make money. They have a business model that involves marketing divisions. Currently, there is a WALL between the Mavic-2 Pro and Phantom 4 Pro V2.0. There is no competition forcing DJI to move that wall...."yet". When that future event happens, than DJI will adjust its marketing model. Until then....well? This is OUR fault for having expectations that were not realistic at this point in time.

I get it now. I see the marketing "bigger picture" that is going on today and Autel could force DJI's hand down the road. But today? Fly your M2P and be happy with HQ and photos. Dont fly FOV if you dont like it. (I hate FOV but can live with HQ)

 

[AeroJ]

There is no competition forcing DJI to move that wall...."yet".

Oh I dunno. That Autel EVO manages 4K60 and very decent video quality (which I'd say holds up pretty well compared to the P4V2) despite not having a 1" type sensor...

 

[Cliff_622]

Oh I dunno. That Autel EVO manages 4K60 and very decent video quality (which I'd say holds up pretty well compared to the P4V2) despite not having a 1" type sensor...

You could be very right on that. I will have an EVO for testing next week! Im VERY curious to see how she looks!