One of the things I was really looking forward to with the P4 was being able to shoot in 1080P at 120fps. When I tested it out on my first flight I noticed that the video looked surprisingly soft, so I ran a few tests today. It seems that at 120fps it only uses part of the sensor, but even when accounting for that the image is very soft.

Best to demonstrate with a few images, which are posted below. All of these were taken from the same spot at ISO100, shutter 1600 (no ND filter yet) and D-Log. All clips, except for the 120fps were at -1 sharpen, -1 contrast and -1 saturation, the 120fps clip was set to standard. Nothing was changed in post, I just added the labels.
If you compare the 1080p30 to the 1080p120 image, you'll see that the 1080p30 shot is much wider. You need to enlarge it by about 200% to have the same frame as the 1080p120 shot. If you take a 4k30 shot and put it into a 1080p timeline and do NOT scale it, you get about the same frame as the 1080p120 shot, except that it is much sharper. I have also included a scaled version of the 4k30 shot for comparison.

To me the 1080p120 shot looks extremely soft. I don't think that is a problem with my camera, but it would be great to hear if someone else has a sharper image shooting in 1080p120 (and I don't mean by increasing the sharpness in camera or in post) or if this is just the way it is. If that's the case, I am hoping that a firmware upgrade might be able to improve that.

Thanks.

Why are they using only part of the sensor?

That's kind of messed up to have not disclosed that you'll be stuck with much wider image areas.  And from these pics, I would agree it seems much softer focus -- disappointing.

citivas Posted at 2016-3-17 19:59
Why are they using only part of the sensor?

That's kind of messed up to have not disclosed that you ...

I'm guessing they probably have to limit processing to achieve the high frame rate, so by only using the central part of the sensor, they're processing less pixels.  They did say somewhere that the field of view reduces when shooting 120 fps, which implies a cropped sensor.

When you resample down from the full sensor resolution, you get a sharper image than when you just crop the centre of the image.

nigelw Posted at 2016-3-18 04:25
I'm guessing they probably have to limit processing to achieve the high frame rate, so by only usi ...

I totally agree with you Nigel, that that's the reason they are using only part of the sensor. It saves them the processing power.

But the weird part is that if you look at the 4k30 clip that is NOT scaled or sampled down to 1080, but just cropped, you still have a much sharper image. That's what I would expect the 1080p120 clip to look like, and frankly, that's what I hope can be fixed.

Thanks for your input.

nebelung Posted at 2016-3-17 20:37
I totally agree with you Nigel, that that's the reason they are using only part of the sensor. It  ...

But 4K frames are still sampled down from a 12mp image to an 8mp image since it uses the whole sensor, which has 12mp.  So the 4k frames will be sharper to start with.

Spec sheet says 120 fps uses half the sensor size that 60 fps uses.  It must then use pixel doubling to get back to the 1920 x 1080 frame size, hence the softer image.

hummingbird.uav Posted at 2016-3-17 21:20
Spec sheet says 120 fps uses half the sensor size that 60 fps uses.  It must then use pixel doubling ...

60 fps uses the full 12mp sensor & resamples down to 1920 x 1080.  Half of the sensor width & height would be a quarter of the pixels, so 3mp, which is still more than 1920 x 1080, so it's still resampling down, but to a lesser degree.

hummingbird.uav Posted at 2016-3-18 05:20
Spec sheet says 120 fps uses half the sensor size that 60 fps uses.  It must then use pixel doubling ...

Could you point me to where it says that in the spec sheet? The one I find online, which matches the manual does not make that distinction: http://www.dji.com/product/phantom-4/info#specs

Thanks.

If they are pixel doubling to get to 1080P, I don't think the result would be as useful as just sticking with 60fps and frame doubling.  I guess it depends on the application but I would take a clearly picture in most cases.

nebelung Posted at 2016-3-17 22:49
Could you point me to where it says that in the spec sheet? The one I find online, which matches t ...

I couldn't find it there either, but the clue is in the FAQ.

13. What makes the Phantom 4 camera better?
The Phantom 4 camera has been improved in terms of image quality. Chromatic aberration has been reduced by 56% and lens distortion had been reduced by 36% compared to Phantom 3 Professional. In addition to lens improvements, camera firmware has been tweaked to enable the camera to capture 120fps video in full 1080p FHD (the camera’s field of view will be reconfigured to 47° when recording in this mode) for smooth slow motion.

47° is half of the normal 94° field of view, which means it's using half the width x half the height of the sensor, or a 2x crop factor.  It's should declare it in the lens or video specs, but they appear to have forgotten.

nigelw Posted at 2016-3-18 04:52
But 4K frames are still sampled down from a 12mp image to an 8mp image since it uses the whole sen ...

I don't think that is 100% correct. Here's my understanding, but it would be great to get an official answer from DJI on this.

The 1/2.3" chip has an effective pixels size of 12MP. That is 4000p x 3000p, a 4:3 image. For a 4K video image that is 16:9, it would simply only use the effective pixels needed. At 4096×2160 it uses the whole width of the sensor, but only 2160p of the full 3000p hight. That gets us the about 8.8MP for 4k. No need to sample it down.

At 1080p30 or 60, it uses the same pixels on the chip, but samples them down to a 1920x1080 image. At 1080p120 it apparently does not sample down the image, but only uses 1920p x 1080p in the center of the chip, hence no need to down sample. But that also would mean that a 1080p120 image should be as sharp as 1920p x 1080p area in the center of a 4k shot.

I am starting to wonder if it could be result of the video bitrate. A 1080p30 has a 40Mbps bitrate and a 1080p60 clip has a bitrate of 60Mbps (which is the max). The 1080p120 clip is output as a 1080p30 video. If you play that clip back at normal speed in Premiere or Quicktime, it is a slow motion movie. That 1080p30 clip has a bitrate of 12.5Mbps.

I get that is can't be 60Mbps for the slow motion clip, because it has to write 4 frames for every frame of the 30fps recording, or 2 frames for every frame of the 60fps recording. But based on the 60fps clip, we should be able to get close to 30Mbps for the 120fps recording, which would be a huge improvement.

Any chance to get an "official" word from @DJI-Amy or @DJI-Tim?

Thanks.

nebelung Posted at 2016-3-17 23:37
I don't think that is 100% correct. Here's my understanding, but it would be great to get an offic ...

Yes, sorry, I forgot about the 16:9 crop.  My head's in photo mode.

"But that also would mean that a 1080p120 image should be as sharp as 1920p x 1080p area in the centre of a 4k shot."

Yes, it should be very close.  Looking at your samples again, it looks like the 4K crop (not scaled) has had more sharpening applied.  Maybe try different sharpening settings & see if you can get the same look.  How do the file sizes compare? That might show you if more compression is being used.  I'd expect extra compression would just show more artefacts though, rather than lower sharpness.

nigelw Posted at 2016-3-18 08:00
Yes, sorry, I forgot about the 16:9 crop.  My head's in photo mode.

"But that also would mean tha ...

I'll try to play with the sharpness, but in the examples that I have posted the 4k30 clip was at -1 sharpness and the 1080p120 was at 0 sharpness.

I'll also shoot 1 minute clips in each mode to compare file sizes, and yes, you are probably right lack of sharpness is not a result of a lower bitrate.

Thanks for trying to help me figure this out. I'll report back after more tests.

nebelung Posted at 2016-3-17 18:06
I'll try to play with the sharpness, but in the examples that I have posted the 4k30 clip was at - ...

I see you have found the reference in the FAQ about how the 120fps image is sampled.
The statements about which part of the sensor are used for 4K and 1080P 60 make sense.  Both use the whole sensor width but the 1080P does not use every pixel of the sensor.  For this reason the 1080P 60 is sharper because there is less interpixel noise.  I would expect the sharpness of 4K and 1060P 120 to be similar because the image is captured using adjacent sensor pixels.  Have to wait until the machine is delivered and test it.

this is what I expected that image quality will go down when using 120

nebelung Posted at 2016-3-18 07:37
I don't think that is 100% correct. Here's my understanding, but it would be great to get an offic ...

Hello Nebelung,

You wrote that "it has to write 4 frames for every frame of the 30fps recording". So i guess that the problem of the sharpness is not a question of the effective sensor size used but maybe just a motion blur created by using 4fps=1fps written.

Hope i am clear enough with my poor english...

patou72 Posted at 2016-3-19 03:16
Hello Nebelung,

You wrote that "it has to write 4 frames for every frame of the 30fps recording".  ...

Hi patou72, I don't think it's motion blur.

The camera was stationary when I was shooting the examples, and usually a higher frame rate means less blur, which is one of the reasons why many people didn't like the 120fps version of The Hobbit.

I ran some more tests, which I'll post below.

Based on the feedback here, I ran some more tests.

1. I recorded about 30 seconds of video in the different record settings to compare the resulting files sizes.

• 1080p30 created a 1080p30 file. Size ~160MB. Bitrate 40 mbits.
• 1080p60 created a 1080p60 file. Size ~240MB. Bitrate 60 mbits.
• 1080p120 created a 1080p30 file. Size ~200MB. Bitrate 12.5 mbits.
• 4k30 created a 4k30 file. Size ~240MB. Bitrate 60 mbits.
  

So the amount of data recorded for the 1080p120 is somewhere between the 1080p30 and 1080p60 file. There might be room for improvement there. I recall that the 1080p60 mode on the Inspire1 was initially only 40 mbits, but they eventually improved it to 60 mbits. But we are once again coming up against that 60 mbit limit.

2. I changed the Sharpness setting in the 1080p120 mode and recorded short clips with each setting. I also recorded a clip in 1080p60 and 4k30 with my usual setting of -1 Sharpness for comparison. I posted the YouTube video below (make sure to watch in 1080).



The closest match to the cropped 4k30 shot (-1 Sharpness) seems to be somewhere +1 and +2 Sharpness in 1080p120. The lower bitrate introduces a number of compression artifacts at the +2 setting, so I'll move forward with a +1 setting for now.

patou72 Posted at 2016-3-18 19:16
Hello Nebelung,

You wrote that "it has to write 4 frames for every frame of the 30fps recording".  ...

The frame rate will limit the maximum shutter speed to 1/120 sec, so blur is much less likely.

nebelung Posted at 2016-3-18 20:27
Based on the feedback here, I ran some more tests.

1. I recorded about 30 seconds of video in the d ...

Sounds about right.  They have to lower the sharpening amount because there's not enough data to avoid compression artefacts becoming ugly.  It seems there's no free lunch.

nigelw Posted at 2016-3-19 05:17
Sounds about right.  They have to lower the sharpening amount because there's not enough data to a ...

Stupid free lunch rule...I'll settle for a free beer at this point.

The good news is that theoretically there should be a little bit of room for improvement even inside of the 60 mbits limit.

I'll see how 1080p120 works out in real world examples...otherwise it's back to 1080p60.

Thanks for your insights.

nebelung Posted at 2016-3-18 21:41
Thanks for your insights.

No worries...I'm learning new stuff as well, coming from photography to video has a learning curve.

nigelw Posted at 2016-3-19 05:10
The frame rate will limit the maximum shutter speed to 1/120 sec, so blur is much less likely.

Hey nigelw,

I understand. But anyway we are also working here with a gimbal (and not with a fixed tripod) with a precision of 0.02°...so there could be some micro mouvements. Remember the gimbal of the inspire1 drifted a lot of time for several degrees whilst the bird was sitting on the ground...the only thing you need for that is a magnetic interference in the compass or value changes in the IMU...

It's possible that i'm completely wrong but the gimbal is not at 100% free of micro corrections

patou72 Posted at 2016-3-18 22:48
Hey nigelw,

I understand. But anyway we are also working here with a gimbal (and not with a fixed  ...

It doesn't matter if the gimbal moves.  A 40mm (35mm format equivalent) lens exposed for 1/120 sec will not generally show motion blur unless the movement is very fast.  I've tested & achieved consistent 2 second exposures without motion blur in windy conditions.  Also, 1/120 sec is the worst case possible.  It's more likely the shutter speed was well in excess of that in good sunlight, probably nearer 1/1000 sec.

citivas Posted at 2016-3-17 15:59
Why are they using only part of the sensor?

That's kind of messed up to have not disclosed that you ...

they did disclose it. They said it would have a 47 degree fov (equivalent to 35mm) when the camera was in 1080p 120fps i believe on the product page, or when one of them on google.

FatedFilmsNC Posted at 2016-3-19 09:02
they did disclose it. They said it would have a 47 degree fov (equivalent to 35mm) when the camera  ...

For what's it worth, this information should not only be in the FAQ, but more importantly in the specs.

But my main concern is really not about the reduced FOV, but how much softer the image is compared to all the other modes. Some of this can be addressed by changing the sharpness settings in the camera, but the (limited) bitrate will probably only allow for some sharpening before you pay a penalty in compression artifacts.

It's less than optimal and I hope that DJI will take another look at this to see if they can improve things a little.

nebelung Posted at 2016-3-18 21:33
For what's it worth, this information should not only be in the FAQ, but more importantly in the s ...

From my experience with DJI, they will most likely see your post and actually look into something to make the feature better with a firmware update in the future. Stay tuned!