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

AroundTheWorld · Feb 6, 2019

kilomikebravo said:
ATW: If you choose not to post-process your videos, then by all means shoot in Auto and 8-bit. But if you really care about the final quality of your work, then post-processing is required and as WTB stated, the subject has been covered extensively. The truth of the matter is you will lose a great deal of image quality if you forego 10-bit DLog and post-processing. Z'up to you, shipmate.

Thanks for sharing your point of view! I will probably try both and compare.

AroundTheWorld · Feb 6, 2019

Sorry guys I am slowly reading all the pages of this thread but I was wondering if we did actually find out what method dji finally used to render the full fov image in 4k?

Is it subsampling / line skipping / pixel binning from the 5.5k sensor?

Or is it an upsampling from a 2,7k image to 4k?

Cause I know dji claimed it was subsampling but many people seemed to question their statement

CanadaDrone · Feb 6, 2019

AroundTheWorld said:
Sorry guys I am slowly reading all the pages of this thread but I was wondering if we did actually find out what method dji finally used to render the full fov image in 4k?

Is it subsampling / line skipping / pixel binning from the 5.5k sensor?

Or is it an upsampling from a 2,7k image to 4k?

Cause I know dji claimed it was subsampling but many people seemed to question their statement

They claim the method they use is subsampling, which is basically pixel skipping. It skips certain pixel groups when reading the sensor, and then it is processed into a 4K image while maintaining full FOV (28mm equivalent). In 4K HQ, it is only reading a 4K portion of the sensor, which is 1:1 pixel readout (of the active area), but the disadvantage is of course the heavy crop (approximately 77 degrees FOV to 55 degrees FOV). Depending on how the subsampling is done, it can also skip lines, so I have not seen anything so far that would make me think DJI's statement is untrue. Both FOV and HQ methods reduce the burden on the processing side, which is clearly what DJI's goal was. It's not a great way to get 4K, but that is the price we pay for size I suppose. Ideally they would be sampling the entire sensor for every frame, and properly processing it into 4K - but that requires more processing power than they apparently have.

AroundTheWorld · Feb 6, 2019

CanadaDrone said:
They claim the method they use is subsampling, which is basically pixel skipping. It skips certain pixel groups when reading the sensor, and then it is processed into a 4K image while maintaining full FOV (28mm equivalent). In 4K HQ, it is only reading a 4K portion of the sensor, which is 1:1 pixel readout (of the active area), but the disadvantage is of course the heavy crop (approximately 77 degrees FOV to 55 degrees FOV). Depending on how the subsampling is done, it can also skip lines, so I have not seen anything so far that would make me think DJI's statement is untrue. Both FOV and HQ methods reduce the burden on the processing side, which is clearly what DJI's goal was. It's not a great way to get 4K, but that is the price we pay for size I suppose. Ideally they would be sampling the entire sensor for every frame, and properly processing it into 4K - but that requires more processing power than they apparently have.

Oh ok great thanks for the details so is that the best method they could have used to render a 4k image from a 5.5k sensor?

At least it's not an upscale from 2.7k resolution just like we were doubting at first

peros550 · Feb 7, 2019

Setting Sharpness to +1 might be of help to some folks who see smoother pictures coming out of MP2.

Just give it a try and you are happy with the result leave it this way.

Ahrimaan · Feb 7, 2019

peros550 said:
Setting Sharpness to +1 might be of help to some folks who see smoother pictures coming out of MP2.

Just give it a try and you are happy with the result leave it this way.

That’s the biggest deal here : People think sharp is better but it isn’t !
The P4P just doing some massive digital sharpness , the M2P don’t . And that’s what some people call „quality“
No it’s just oversharpening !

CanadaDrone · Feb 7, 2019

AroundTheWorld said:
Oh ok great thanks for the details so is that the best method they could have used to render a 4k image from a 5.5k sensor?

At least it's not an upscale from 2.7k resolution just like we were doubting at first

They claim due to processing power constraints, that subsampling was the best method to give us the maximum quality while reducing the burden on the processing side enough to allow continuous 4K recording (other methods they could have used to achieve a similar result would be line skipping or pixel binning). That is probably true, and whether or not they could have easily overcome that is something we'll never know, or maybe they are saving it for the Platinum. From a marketing perspective, they less they can deliver while still creating a massive demand, the better - not saying they necessarily did anything on purpose, just generally speaking.

I think where it gets more confusing is that DJI says the entire 16:9 area of the sensor is sampled, but the definition of subsampling suggests that is not true (skipping pixel groups), so there is some disconnect there, or at least requires further explanation from DJI. ~5.5K video would be approximately the full width of the 20MP sensor, but vertically trimmed for 16:9 or 1.78:1. I suspect they are just trying to make it sound slightly better than it is, which isn't surprising of any "official response" from any company.

At least the video quality is still quite good, it's just not amazing. I think the sweet spot for a compromise would have been HQ-mode-quality video with full FOV, but maybe next time. For me the most disappointing thing is that DLog-M does not correct for distortion, but that is more of a personal annoyance due to the editing software I like to use (LumaFusion) not yet supporting lens corrections. Someone coming form a Mavic Pro or Mavic Air though is still going to see an enormous upgrade especially with graded 10bit footage. The Mavic 2 Zoom does not offer 10bit at all, but isn't a bad choice for someone that only shoots video and someone that isn't fond of post processing.

That's just my take on it anyway, I would like more clarification from DJI but I don't know if we'll get it.

paulatkin73 · Feb 7, 2019

CanadaDrone said:
They claim due to processing power constraints, that subsampling was the best method to give us the maximum quality while reducing the burden on the processing side

People keep talking about that sub-sampling as if it a something bad or non-desirable, while, it is not at all. It is used on every single device with a sensor that is bigger than a standard 4K resolution matrix, which is 3840x2160. Only part that differs is the power of post-processing method used when this subsampling is executed, how exactly data merges back, from how many source pixels, and how fast is scans the whole matrix.

Plus, i think the bottom line is - anybody who is not happy of how mavic works should simply go ahead and buy something from the zenmuse family. or start their own business and produce a gimbal that will cost less than the one DJI made, but with better specs and a better sensor. I bet it is doable. but to b$tch non stop that a zenmuse capability was not pressed into mavics form factor is really not productive - and it seems no matter how many times folks like you or i would try to explain it to those folks - it is not working. my approach is simple - do not like it - do not use it. or build something better yourself.

AroundTheWorld · Feb 7, 2019

I

CanadaDrone said:
They claim due to processing power constraints, that subsampling was the best method to give us the maximum quality while reducing the burden on the processing side enough to allow continuous 4K recording (other methods they could have used to achieve a similar result would be line skipping or pixel binning). That is probably true, and whether or not they could have easily overcome that is something we'll never know, or maybe they are saving it for the Platinum. From a marketing perspective, they less they can deliver while still creating a massive demand, the better - not saying they necessarily did anything on purpose, just generally speaking.

I think where it gets more confusing is that DJI says the entire 16:9 area of the sensor is sampled, but the definition of subsampling suggests that is not true (skipping pixel groups), so there is some disconnect there, or at least requires further explanation from DJI. ~5.5K video would be approximately the full width of the 20MP sensor, but vertically trimmed for 16:9 or 1.78:1. I suspect they are just trying to make it sound slightly better than it is, which isn't surprising of any "official response" from any company.

At least the video quality is still quite good, it's just not amazing. I think the sweet spot for a compromise would have been HQ-mode-quality video with full FOV, but maybe next time. For me the most disappointing thing is that DLog-M does not correct for distortion, but that is more of a personal annoyance due to the editing software I like to use (LumaFusion) not yet supporting lens corrections. Someone coming form a Mavic Pro or Mavic Air though is still going to see an enormous upgrade especially with graded 10bit footage. The Mavic 2 Zoom does not offer 10bit at all, but isn't a bad choice for someone that only shoots video and someone that isn't fond of post processing.

That's just my take on it anyway, I would like more clarification from DJI but I don't know if we'll get it.

You are right HQ mode quality with full FOV would have been amazing. But it would have implied a subsampling method anyway. I'm just wondering how they were able to achieve this crystal crisp image on the Phantom 4 Pro considering that they probably used the same subsampling method.

If the processor is in cause it's quite a shame cause It wouldn't have bothered me to pay more

CanadaDrone · Feb 7, 2019

AroundTheWorld said:
I

You are right HQ mode quality with full FOV would have been amazing. But it would have implied a subsampling method anyway. I'm just wondering how they were able to achieve this crystal crisp image on the Phantom 4 Pro considering that they probably used the same subsampling method.

If the processor is in cause it's quite a shame cause It wouldn't have bothered me to pay more

There is a difference between sampling the entire sensor, and then downsampling the full sensor data from each frame to 4K compared to not recording all of that data in the first place (pixel skipping, line skipping, etc.) The former yields much better quality. The better your input data, the better your output data, assuming adequate processing. A similar example would be if I took a 46MP image with my Nikon D850 and downsampled it to 11MP, it will always be better than if the camera only recorded ~1/4 of the pixels in the first place and then combined that data into a 11MP image (all else equal). You get numerous benefits including noise reduction, artifact reduction, less aliasing, etc. If you have enough resolution, you don't have to use the whole sensor to get the same benefits (I.e. downsample 6K to 4K like Sony does), but then there will be a crop factor involved. Taking this one step further, 4K properly downsampled to 1080P is better than native 1080P all else equal.

I'm not as familiar with the P4P's video so I don't know what they are doing there, but there is likely more power available as it will do 60fps and I am pretty sure it also corrects for distortion in DLog, which the M2P does not. It is only 8bit though to my knowledge. So far anyway, it looks like DJI is using a mechanical shutter and 60fps as the primary differentiation between the Mavic and Phantom line - otherwise there would be little reason not to just get the M2P. The P4P also is locked to ISO 500 in DLog.

I am with you on your last point - I would have paid more without even thinking about it for 'proper' 4K if it was a matter of processor cost, etc. but it's not like we have any other good options if you want a foldable/portable drone with a larger than average sensor so we're at the mercy of whatever DJI does.

markyga · Feb 9, 2019

Hey dude,

Great video... very technical.

So here’s my question... what about still shots? Specifically the 20MP camera. How does it compare to the Phantom Pro?

Blue skies,
M.S.

WithTheBirds · Feb 9, 2019

The an

markyga said:
Hey dude,

Great video... very technical.

So here’s my question... what about still shots? Specifically the 20MP camera. How does it compare to the Phantom Pro?

Blue skies,
M.S.

Your question has been addressed in the thread- quite exhaustively.

maven4336 · Mar 1, 2019

markyga said:
Hey dude,

Great video... very technical.

So here’s my question... what about still shots? Specifically the 20MP camera. How does it compare to the Phantom Pro?

Blue skies,
M.S.

I have both and hard to say colors look better on Mavic2pro but phantom 4pro just has its own look on photos, I fly both every day and both are killer in there own way

AroundTheWorld · Mar 31, 2019

One of the best video, very technical, comparing the image quality of the Full FOV mode and HQ mode of the Mavic 2 Pro versus the Phantom 4 Pro

Unfortunately its in French