I have a Phantom 4 pro a few months ago already, but almost since the beginning I was not too much impressed with the quality obtained, specially with the 4K videos, now I think that could be due to a technical feature that I wish to be very very wrong about it.

In these 3 100% crops are something in common: signals of interpolation or deinterlacing, as you can see in the stairs like artifacts on each diagonal and line captured by the P4P camera, this kind of artifacts can be found on deinterlaced footage converted to progressive or in FullHD upscaled to 4k:


so... P4P camera is really 4K? I have another 4K cameras and none of their footage presents this kind of artifacts.

These are the details of the video where the crops were taken:
https://www.dropbox.com/s/ovstiwrj3cchjhr/video_details.jpg?dl=0

Is the video from the SD card or the mobile device?


EDIT: ignore that, I missed the dimensions on the clip

Yup, it's 4K.  It's just not the best.  

I'm not sure what the situation is with your P4P, but with mine, it shoots exceptional 4k.  It makes 1080p look way lower quality.

It doesn't have a 4K image sensor, the sensor is bigger than 4K but not a nice 2x so a perfect scaling is not possible.

Most of your image looks like decent 4K though, maybe a little out of focus on the closer bits of the bridge.

I wonder if the issue you are seeing is a result of stabilisation trying to remove vibrations?
Are your propellers well balanced or do they vibrate a lot?

What sharpness settings are you using?

up to 4k 30fps the quality is quite good and better than most small sensor cameras e.g. gopro, mavic, phones but above 30fps p4p is doing line skipping so it introduces aliaising and you loose vertical resolution.... try filming at home without flying a test chart or news paper at 4k 30 and 60 and compare.

Nigel_ Posted at 2017-9-14 07:59
It doesn't have a 4K image sensor, the sensor is bigger than 4K but not a nice 2x so a perfect scaling is not possible.

Most of your image looks like decent 4K though, maybe a little out of focus on the closer bits of the bridge.

Well I was sure the conditions were as stable as P4P can offer, and in this matter it does a very good job, it seems like line skipping or a kind of de-interlacing.

gt3rs Posted at 2017-9-14 12:10
up to 4k 30fps the quality is quite good and better than most small sensor cameras e.g. gopro, mavic, phones but above 30fps p4p is doing line skipping so it introduces aliaising and you loose vertical resolution.... try filming at home without flying a test chart or news paper at 4k 30 and 60 and compare.

Thanks a lot for your comment, and well, I think you are right, I've compared with another 4K@30fps footage and I couldn't found the same issue, so certainly is line skipping, perhaps to achieve the higher frame rate, thus, the conclusion could be that P4P is not a real 4K/UHD at 60fps, that is a quite bad news because must of the cases people shoot at 60fps for slow motion where the artifacts became even more visible. This is a huge negative point for P4P.

fansde3bbad0 Posted at 2017-9-14 21:08
Thanks a lot for your comment, and well, I think you are right, I've compared with another 4K@30fps footage and I couldn't found the same issue, so certainly is line skipping, perhaps to achieve the higher frame rate, thus, the conclusion could be that P4P is not a real 4K/UHD at 60fps, that is a quite bad news because must of the cases people shoot at 60fps for slow motion where the artifacts became even more visible. This is a huge negative point for P4P.

It is not line skipping. To do that the sensor must have a vertical resolution at least equal to twice the target vertical resolution, otheriwse you would see horendous aberrations in the picture. The answer is shown in the properties window of the OP's grab. Look at the properties.

Nigel_ Posted at 2017-9-14 07:59
It doesn't have a 4K image sensor, the sensor is bigger than 4K but not a nice 2x so a perfect scaling is not possible.

Most of your image looks like decent 4K though, maybe a little out of focus on the closer bits of the bridge.

You might be on the right track with the stabilization and the scaling down.

Geebax Posted at 2017-9-14 21:52
It is not line skipping. To do that the sensor must have a vertical resolution at least equal to twice the target vertical resolution, otheriwse you would see horendous aberrations in the picture. The answer is shown in the properties window of the OP's grab. Look at the properties.

I'm missing it I guess.

If you could upload an original 4k video from the SD card to the Dropbox, then provide me a link to download it, I'll forward it to our engineer for analysis.

The P4p sensor is the same as in the Sony RX100m4, so we know a fair bit about it. So YES, it is true 4K. Quite good 4K, actually!

However, the main point to keep in mind is: you cannot compare a video still to a JPEG photo, in case that is what you did...  it will always look considerably less sharp  (that goes for practically all cameras)

Why?

image processors use shortcuts to handle the massive amount of image data coming off sensors.
1. Pixel binning / line skipping = not all pixels are used to compile the image. P4p has 20MP sensor and will only use a portion of the total 20mio pixels to create an 8MP = 4K image frame. this results in significant loss of sharpness, moire and aliasing (compared to a native 8MP photo).
2. Chroma Subsampling = color information from 4 pixels are combined and averaged into 1. that results in noticeable loss in clarity.

bottomline.. on basically every camera, video frames will look nowhere near as good as photos off the same sensor, even at same resolution.
what other 4K cameras do you have / are you comparing?

disclaimer: half-baked amateur talking here . ..

Sebb Posted at 2017-9-15 01:56
The P4p sensor is the same as in the Sony RX100m4, so we know a fair bit about it. So YES, it is true 4K. Quite good 4K, actually!

However, the main point to keep in mind is: you cannot compare a video still to a JPEG photo, in case that is what you did...  it will always look considerably less sharp  (that goes for practically all cameras)

One of the reasons I decided for P4P was just Sony is using the same sensor.

The crops were extracted from a 4k@60fps video, aren't stills, they were encoded carefully to avoid another artifacts different to the mentioned one in this post, the issue is clearly regular, is not due to compression, or something similar, we can see an ordered pattern just one scanline after another, like the ones we can see in a de-interlaced or line-skipped video frame.

I can confirm the issue is only present in 4k@60fps videos not in 4k@30fps, so must be something associated with the framerate.

DJI Diana Posted at 2017-9-15 01:26
If you could upload an original 4k video from the SD card to the Dropbox, then provide me a link to download it, I'll forward it to our engineer for analysis.

Hi,

Well, the video I use for the crops is quite long and big, so I uploaded a couple of short ones where the issue can be noted too:


Video Link:
http://3dlogical.com/_temp/dji/DJI_0061.MOV

You can note the issue on the white lines over the field


Video Link:
http://3dlogical.com/_temp/dji/DJI_0018.MOV

You can note the issue looking at the mountains silhouette.

These videos are unprocessed, unedited, untouched

Could it be something to do with SD Card write speed ?

lannes Posted at 2017-9-16 22:40
Could it be something to do with SD Card write speed ?

The SD card has nothing whatsoever to do with the image quality. It is a digital recording, not analog.

fansde3bbad0 Posted at 2017-9-16 14:11
Hi,

Well, the video I use for the crops is quite long and big, so I uploaded a couple of short ones where the issue can be noted too:

'These videos are unprocessed, unedited, untouched'

The videos might be, but the crops certainly are not. You cannot expect to crop an image like that without creating distortions. The short clip you linked to looks perfect, why are you pixel-peeping?

fansde3bbad0 Posted at 2017-9-16 13:34
One of the reasons I decided for P4P was just Sony is using the same sensor.

The crops were extracted from a 4k@60fps video, aren't stills, they were encoded carefully to avoid another artifacts different to the mentioned one in this post, the issue is clearly regular, is not due to compression, or something similar, we can see an ordered pattern just one scanline after another, like the ones we can see in a de-interlaced or line-skipped video frame.

yes of course high framerates always results in significantly decreased image clarity.
4K60 on P4pro is usable but nowhere near 4K30.

fansde3bbad0 Posted at 2017-9-16 14:11
Hi,

Well, the video I use for the crops is quite long and big, so I uploaded a couple of short ones where the issue can be noted too:

Thanks, I've forwarded them to our engineers, will keep you updated.

The sensor of P4P has about 5470 pixels horizontally and the 4K video is down-scaled from the 5.4K RAW, thus making it sharper and less noisy than directly cropping a 4K window on the sensor.

Unlike 4K 30fps which has square pixels, the 4K 60fps footage is generated from 5.4K RAW with rectangular logical pixels. This technique allows to make full use of the same area of the sensor as that in 4K 30fps mode and keep the full FOV, without sacrificing dynamic range (compared to 4K windowed readout). It is beyond the performance limit of the sensor to readout 5.4K square pixels at a rate of 60fps.

Geebax Posted at 2017-9-16 23:17
'These videos are unprocessed, unedited, untouched'

The videos might be, but the crops certainly are not. You cannot expect to crop an image like that without creating distortions. The short clip you linked to looks perfect, why are you pixel-peeping?

You must have in mind what 100% crop means, they are unscaled 320x320, undistorted, unprocessed images extracted directly from a video frame, and uncompressed because were stored in PNG, so there are no chances to incorporate distortion of any kind from anywhere, any artifact that you could see in those crops actually exists in the video. To obtain unaltered crops of a video I use AfterEffects or Photoshop TimeLine, the video frame is saved directly from a selected frame on the timeline in PNG format, cropped and stored in PNG again, there is no chance that in such as process some distortion, compression artifact or visual contamination be incorporated. But with the video links I provided you can do the crops by yourself and see what I'm seen by your own.

fansde3bbad0 Posted at 2017-9-18 08:27
You must have in mind what 100% crop means, they are unscaled 320x320, undistorted, unprocessed images extracted directly from a video frame, and uncompressed because were stored in PNG, so there are no chances to incorporate distortion of any kind from anywhere, any artifact that you could see in those crops actually exists in the video. To obtain unaltered crops of a video I use AfterEffects or Photoshop TimeLine, the video frame is saved directly from a selected frame on the timeline in PNG format, cropped and stored in PNG again, there is no chance that in such as process some distortion, compression artifact or visual contamination be incorporated. But with the video links I provided you can do the crops by yourself and see what I'm seen by your own.

This is a screen recording showing the process I did to extract 100% crops and compare the differences between 4K@60fps and 4K@30fps of the same subject with the same lighting and ambient conditions, I drop here the links of the videos used for the analysis and the 100% crops extracted from them just made at my house:

Process:
http://3dlogical.com/_temp/dji/ProcessCapture.mp4

4K video at 60fps:
http://3dlogical.com/_temp/dji/DJI_0344.MOV

4K video at 30fps:
http://3dlogical.com/_temp/dji/DJI_0345.MOV

100% Crop 1 at 60fps:
http://3dlogical.com/_temp/dji/Crop@60fps.png

100% Crop 2 at 30fps:
http://3dlogical.com/_temp/dji/Crop@30fps.png

So I hope this is much more clear and generate less confusion  about the cropping and extracting process..

You may have a defective unit, my p4 advanced shots great 4k

fansde3bbad0 Posted at 2017-9-16 14:11
Hi,

Well, the video I use for the crops is quite long and big, so I uploaded a couple of short ones where the issue can be noted too:

Hi, our engineers replied you in the post 22#, please check it.

charlesetischer Posted at 2017-9-18 05:59
The sensor of P4P has about 5470 pixels horizontally and the 4K video is down-scaled from the 5.4K RAW, thus making it sharper and less noisy than directly cropping a 4K window on the sensor.

Unlike 4K 30fps which has square pixels, the 4K 60fps footage is generated from 5.4K RAW with rectangular logical pixels. This technique allows to make full use of the same area of the sensor as that in 4K 30fps mode and keep the full FOV, without sacrificing dynamic range (compared to 4K windowed readout). It is beyond the performance limit of the sensor to readout 5.4K square pixels at a rate of 60fps.

i m not sure i understand fully what you re describing, however we do know that
- sensor can readout full resolution (20.2MP) at maximum 25fps = using all pixels
- sensor can readout cinema 4K resolution (9MP) at maximum 60fps = using partial scan

AFAIK actual framerate limitations of the camera are always introduced by the picture processer - the component that handles RAW uncompressed data frames and encodes them into compressed frames at a certain bitrate, encapsulated into a container. The RX100 m4/m5 for example cannot produce 4K60 frames, despite the sensor having such a readout mode. The Bionx processor can only effectively handle 30fps and already overheats after 5min as it is.

regarding different image quality, I highly doubt that 24/30fps video is using full readout mode (which doesnt even do 30fps), and only 60fps uses partial scan mode.

Today only few consumer cameras actually do 4K with full readout, since it requires siginificant processing. and since it is a premium feature, it is immediately included into marketing. See for example A6500 "4K movie recording with full pixel readout/no pixel binning". We dont see this with P4pro. RX100 also lack it which is expected at this price point.

having said that, all 4K modes with P4pro come from 9MP / 4.5K readout which is partial scan which has to be enlarged again, hence softer image. My assumption then is that 24/30fps can be compiled at full quality into quite clean 4K, while 60fps sacrifices yet additional pixels or lines, resulting in noticably softer footage.

not sure where you got your information, but please share any additional insights.

sorry for long post

Sebb Posted at 2017-9-25 06:56
i m not sure i understand fully what you re describing, however we do know that
- sensor can readout full resolution (20.2MP) at maximum 25fps = using all pixels
- sensor can readout cinema 4K resolution (9MP) at maximum 60fps = using partial scan

That is not correct. No modes of P4P come from 9MP/4.5K readout. Everything at or below 60fps is from 5.4k horizontal readout.

The P4P's 30fps video formats all come down from 5.4k (about 16.8M pixels), using the full width of the sensor. The 60fps modes also use 5.4K horizontally, but each logical pixel is not square, so the minimum resolvable detail in the vertical direction is not as good as in the horizontal direction. However, the readout window still covers the same area as the 30fps modes (using the full width of the sensor), so the FOV does not change and we do not waste the noise performance of the Type 1'' sensor.

It is impossible for a 16:9 movie to use the full 3:2 sensor. But we regard it as full sensor readout as long as the full width is used.

DJI_camera Posted at 2017-9-26 01:54
That is not correct. No modes of P4P come from 9MP/4.5K readout. Everything at or below 60fps is from 5.4k horizontal readout.

The P4P's 30fps video formats all come down from 5.4k (about 16.8M pixels), using the full width of the sensor. The 60fps modes also use 5.4K horizontally, but each logical pixel is not square, so the minimum resolvable detail in the vertical direction is not as good as in the horizontal direction. However, the readout window still covers the same area as the 30fps modes (using the full width of the sensor), so the FOV does not change and we do not waste the noise performance of the Type 1'' sensor.

thank you for clearing that up!
I just saw RX100m4/5 also have full readout. I did not expect that to be the case either - quite incredible.

> Everything at or below 60fps is from 5.4k horizontal readout.
how about higher fps modes, e.g. 1080p120?
I havent tested it on P4pro, but the Rx100m4 has really clean 120fps for such a small camera -> from last weekend

either way - hats off to you guys implementing that sensor in this platform .. i m impressed every time i view my flight footage, especially when comparing to earlier Phantoms and Mavic.

Sebb Posted at 2017-9-15 01:56
The P4p sensor is the same as in the Sony RX100m4, so we know a fair bit about it. So YES, it is true 4K. Quite good 4K, actually!

However, the main point to keep in mind is: you cannot compare a video still to a JPEG photo, in case that is what you did...  it will always look considerably less sharp  (that goes for practically all cameras)

Surprisingly a 4K frame grab from P4A/P4P with a fast shutter is almost exactly the same in regards to detail and sharpness as an out of camera JPG with all profile settings equal and drone stationary/hovering.

charlesetischer Posted at 2017-9-18 05:59
The sensor of P4P has about 5470 pixels horizontally and the 4K video is down-scaled from the 5.4K RAW, thus making it sharper and less noisy than directly cropping a 4K window on the sensor.

Unlike 4K 30fps which has square pixels, the 4K 60fps footage is generated from 5.4K RAW with rectangular logical pixels. This technique allows to make full use of the same area of the sensor as that in 4K 30fps mode and keep the full FOV, without sacrificing dynamic range (compared to 4K windowed readout). It is beyond the performance limit of the sensor to readout 5.4K square pixels at a rate of 60fps.

The P4A/P4P 4K frame is actually a crop of the 5470 sensor pixels after lens correction has been applied. The 4K frame is taken from about a 13-14MP crop of the full 20mp sensor. It is not down scaled from the sensor width of 5470 (it is lens corrected and then cropped from the middle section of the corrected image). So yes the video is using the full sensor (not a cropped sensor readout) but it is not scaling that whole sensor image down to get 4K (just taking a square pixel 4K crop of the 5470 after lens correction).

My personal tests... settings: Normal and Soft (-1+0+0)

Note just my theory. Really wanted to find which mode would work best for my needs. But this is only half the picture. Another test needs to be done for detail while moving (does 4K retain the detail vs 1920 once moving across a surface or detailed scenes) and does H265 afford more bit rate / detail over H264 in such cases (in other words is using H265 effectively 2x or 1.5x the bit rate of H264 when both at 100Mbps.

4096: Unless dimension is required, has no advantage in post (panning) as its not using wider sensor crop. Horizontal field of view is the same as 3840 but less vertical FOV. Less bit quality vs 3840.

3840: For a stationary hover scene like in this test (3m height and subjects from 10m to infinity) 3840 produces the best quality image and FOV overall @ 24/30

2720-1920: All modes are taken from about the 4K resolution (or from about a 13-14MP or less crop of the lens corrected frame off the 20mp sensor). Lower resolutions than 4K are scaled down (but not well IMO). More aliasing is introduced with each downscale in size (over sharpening effect in resizing down). The 3840 footage down scaled in PP CC onto a 1920 timeline looks much cleaner (anti aliasing) than native 1920x1080. Detail is roughly the same to my eyes on monitor or 4K TV. To make the 3840 look as sharp and aliased as the native 1920 footage requires +50 sharpening in Lumetri (and it still looks better than native but not by much - both start to show aliasing).

DJI: A sensor crop mode would be nice also if it makes 60fps/120fps easier to work with rather than using the full sensor width to start with. An option in go4 to use crop or full sensor would be prefect. Also I do not understand why it can process 4K at 30fps well but can not process  1920 @ 60/120 well. Guess it must have some limitation in the process or can at least 1920 60gfps be improved upon (old GoPro Hero 3 can do 1920 @ 60 so surely my phantom can so the same without the skipping/binning or whatever is going on to make it so jaggy and blurry

Dingoz Posted at 2017-9-27 04:22
Note just my theory. Really wanted to find which mode would work best for my needs. But this is only half the picture. Another test needs to be done for detail while moving (does 4K retain the detail vs 1920 once moving across a surface or detailed scenes) and does H265 afford more bit rate / detail over H264 in such cases (in other words is using H265 effectively 2x or 1.5x the bit rate of H264 when both at 100Mbps.

4096: Unless dimensions is required has no advantage in post (panning) as its not using wider sensor crop. Horizontal field of view is the same as 3840 but less vertical FOV. Less bit quality vs 3840.

very interestinig, thanks for sharing your test

couple thoughts

just to confirm: you re saying the c4k is basically 4k lettboxed and repacked at c4k resolution? of course i was expecting to 'see' a bit more... but then cropping 4K into c4K would be more or less identical.

i ve managed to get an opinion from wolfcrow on youtube about h.265 vs. h.264. he maintains that
- you really only benefit at low bitrates. where h.264 will break down, h.265 will still work. but 100Mbps  is a decent amount for a clean h.264 picture, therefore not much to be gained.
- h.265 compresses colors far more than h.264, therefore could result in banding and be even less robust in post than 8bit as is
- he would never use it, even on GH5. i hope i did not misquote his reasoning,,

Best
Sebb

Sebb Posted at 2017-9-27 04:37
very interestinig, thanks for sharing your test

couple thoughts

Yes. It's letter boxed. Same effect as taking a larger original image in LR and applying a 4096x2160 vs 3840x2160 aspect ratio crop (horizontal FOV is almost the same but vertical FOV is a fraction more in 4K vs C4K. It also 'appears' that C4K is ever so slightly scaled up vs 4K or is the 4K ever so slighly scaled down more than C4K (could be wrong on this part).

Dingoz,

Thanks for posting that table!  Good stuff. I tend to like using H.264 more often on max resolution/30fps.  I do like to use H.265 and 60fps when I am moving around fast. That extra frame rate helps, though I do notice the slight color degradation.


Steve
Flight distance : 5852
Flight distance + Litchi: 72741
2017-09-24 - Breezy Island Flight

For those interested this is how the video and images are processed into final outputs (approximately):

Without Text Overlays

Original 20MP sensor + lens output (Raw)