Does Mavic Shoot Video or Watercolor Paintings? Don’t settle. Get Involved!

[Dale Rodgers]

This video gives a good overview of why D-log is bad for these cameras - its a bit long winded but the histogram work from 10 minutes on shows how the DJI implementation of D-Log does not increase dynamic range:

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

The thing that drove it home for me was the blown highlight exists in D-log a third of the way from the right - one would expect a highlight to be crushed right up against the right.

As an example, if D-Log is only using half the available range then it stands to reason the image is half as constrasty - too many similar shadesget compressed into one mushy mess by the codec. Lowering contrast to -1 makes the issue even worse and I would have also thought decreasing saturation may make things worse (too many similar tones to compress away). Then if the noise reduction kicks in things go from bad to worse.

I'd expect colour grading from a reduced dynamic range to give "quantisation" after contrast/saturation have been increased in post - this is why the original Mavic D-Log was deemed probably deemed useless.

I've migrated from a P3P to the Mavic and have believed for the last 5 months that the Mavic has a flawed camera or mine was broken (I was very happy with the P3P. I was getting mush one day and okay results the next. However, as I was on the waiting list for so long I was listening to a lot of Youtube clips recommending camera settings and blindly started using them from day 1 of receiving my Mavic. I'm not saying the bloggers are incorrect (those settings may well work for them with their lighting conditions) but a typical British winter has its own "unique" lighting conditions. I'm not saying that the mush isn't a real problem for some but I was starting to believe my Mavic was broken without understanding what I was doing with the settings.

I've been shooting 1080@25fps with NONE(0,0,0) for the last couple of weeks with satisfactory results and very little mush. 2.7K@25fps also is good enough for me. Just try to compose and expose your shots to the best of your abilities at the time of recording. Experiment, experiment, experiment and if in doubt then leave it with the default the manufacturer ships it with.

Incidentally, I was shooting with the Vivid profile on my P3P for ages before I started colour grading and was happy (for me) with the results. I looked at my first early videos with the P3P and have noticed the same artifacts as with the Mavic (I got my P3P in the winter too!).

Good luck

 

[AyeYo]

Hello, based on your qualified experience, tell me what resolution and settings do you use for mavic shooting ?
Personally when I make a 360 degrees, I think 30 fps is too less.

Unfortunately you don't have a choice, because going above 30fps on 2.7k and 4k isn't possible and above 30fps on 1080p destroys the video (something DJI has no intention of fixing).

The trick to getting 360 yaw shots at 30fps is yawing SLOWLY - like, really slowly.

My normal shooting settings are 2.7k, 30fps, None, 0,0,0.

 

[AyeYo]

This video gives a good overview of why D-log is bad for these cameras - its a bit long winded but the histogram work from 10 minutes on shows how the DJI implementation of D-Log does not increase dynamic range:

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

The thing that drove it home for me was the blown highlight exists in D-log a third of the way from the right - one would expect a highlight to be crushed right up against the right.

As an example, if D-Log is only using half the available range then it stands to reason the image is half as constrasty - too many similar shadesget compressed into one mushy mess by the codec. Lowering contrast to -1 makes the issue even worse and I would have also thought decreasing saturation may make things worse (too many similar tones to compress away). Then if the noise reduction kicks in things go from bad to worse.

I'd expect colour grading from a reduced dynamic range to give "quantisation" after contrast/saturation have been increased in post - this is why the original Mavic D-Log was deemed probably deemed useless.

I've migrated from a P3P to the Mavic and have believed for the last 5 months that the Mavic has a flawed camera or mine was broken (I was very happy with the P3P. I was getting mush one day and okay results the next. However, as I was on the waiting list for so long I was listening to a lot of Youtube clips recommending camera settings and blindly started using them from day 1 of receiving my Mavic. I'm not saying the bloggers are incorrect (those settings may well work for them with their lighting conditions) but a typical British winter has its own "unique" lighting conditions. I'm not saying that the mush isn't a real problem for some but I was starting to believe my Mavic was broken without understanding what I was doing with the settings.

I've been shooting 1080@25fps with NONE(0,0,0) for the last couple of weeks with satisfactory results and very little mush. 2.7K@25fps also is good enough for me. Just try to compose and expose your shots to the best of your abilities at the time of recording. Experiment, experiment, experiment and if in doubt then leave it with the default the manufacturer ships it with.

Incidentally, I was shooting with the Vivid profile on my P3P for ages before I started colour grading and was happy (for me) with the results. I looked at my first early videos with the P3P and have noticed the same artifacts as with the Mavic (I got my P3P in the winter too!).

Good luck

Best post of the thread. This should be sticky'd above the first post.

 

[Dale Rodgers]

Unfortunately you don't have a choice, because going above 30fps on 2.7k and 4k isn't possible and above 30fps on 1080p destroys the video (something DJI has no intention of fixing).

The trick to getting 360 yaw shots at 30fps is yawing SLOWLY - like, really slowly.

My normal shooting settings are 2.7k, 30fps, None, 0,0,0.

Tripod mode is a big help keeping a smooth slow yaw.

 

[bucefalo]

little note to this thread..
honestly shooting at MANUAL and adjusting the shutter ( so by day with an iso of 100) seems to give THE BEST results.. i had some doubts about it but honestly is really the best !
someone over a youtube video suggested to LOOK at the value of shutter on "auto " then change to MANUAL , go to the same shutter that mavic has choosen in AUTO and then you can slightly increase or decrease that value , considering your taste/histogram!! today i tried it and yes it was just perfect!

[EDIT] this is the youtube video(at 4:02):

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[Davey]

Thanks for that advice, I think I'll try ART and NORMAL next time ,perhaps with just -1 on contrast and see how it works, I'll keep you updated with results

For me the art setting has worked best with everything dialled to zero and Very very careful focussing on the part of the image that is inclined to get crushed, which is daft because I end up shooting a scene, analysing it on a large monitor and then reshooting if it's not good enough the first time, not a good solution.
I just saw a recent brilliant post recommending cine like profile which I will try but fear the limited dynamic range will still be a problem

 

[MavicPro83]

Best post of the thread. This should be sticky'd above the first post.

Clearly some great points rasied in this vid, I wonder if the drone being used is a P4 or P4P looking at the one in the background so would be interesting to see if the same logic applies to the Mavic as the camera itself likely affects the dynamic range etc too... overall though a good overview for sure.

 

[neo_xiii]

For me the art setting has worked best with everything dialled to zero and Very very careful focussing on the part of the image that is inclined to get crushed, which is daft because I end up shooting a scene, analysing it on a large monitor and then reshooting if it's not good enough the first time, not a good solution.
I just saw a recent brilliant post recommending cine like profile which I will try but fear the limited dynamic range will still be a problem

Hey there ,
Just wondering where did you see the cinelike had a limited dynamic range? from what I saw here and from my personal experience , its dynamic range is pretty decent, and sometimes better than d-log, surprisingly.
cheers!

 

[Realist123456]

So I think I'm honing in a little more on what the capabilities of the Mavic are, along with its strengths and weaknesses.
I went out today and put into practice a few things I've picked up recently and here are my thoughts:
Firstly, after watching Drone Film Guide's piece about d-log vs cinelike a few weeks ago, I have decided to ditch d-log.
Even after the recent update, personally, I just don't think it's worth the trouble.
If the mavic had a sensor that could take advantage of it then fine, but as it's such a crap sensor using d-log and expecting a big improvement is like the proverbial lipstick on a pig.
So I've gone with D-Cinelike. It seems a bit closer to reality and doesn't have as far to push to get to a useable result (IMO).
I've also tested shooting with different sharpness settings and at least for my Mavic, I will be sticking with None 000.
I tested it in bright sunshine today (2160P @ 25fps + 1/50th @100asa +32ND). Exposed very much to the right. It could pull down detail in the scene except with breaking surf - the whites were clipped and the brightest highlites in the white water were gone.
Otherwise the results were quite good with little or no watercolor. Maybe just a tiny hint in some very detailed areas but not enough that you'd notice unless you really went looking for it at 100%.
The footage did have a "sharpness" and "digital" look to it but it wasn't (too) horrendous and I think that's a price I'm prepared to pay to avoid watercolour. Essentially, sharpness is the lesser of two evils.
I went back this afternoon after sundown and shot more footage.
I also did a stills shot.
I've taken screenshots from both and attached them.
The 12MP still is obviously much better than the 4K video but that's a result of compressing 25 frames a second instead of one.
However the screenshot also shows the inherent weakness of the Mavic sensor and/or processing when it comes to video in low light.
Even at a setting of 000 the mavic can't avoid giving quite pronounced watercolour.
That same scene at midday was quite good. But with soft lighting and exposing a bit to the left it just dissolves into watercolor.
So I think the takeaway for me is: D-cinelike. 100asa. Expose as far the right as possible. There's no escape from watercolour if you're shooting a darkish scene.

 

[Martspid]

I am new with the Drone and video stuff, but I have noticed that in one of my videos that the image in some part of my video is good but like in the corner they look blurry.
should I contact DJI about this or there is an alternative solution. Ion the video I tried to focus a few times but still the image is not as expected.
Thank you for your help

 

[newnan3]

Does the d-log and cinelike settings do anything for pictures? Im a photog and I shoot exclusively in raw....

 

[Realist123456]

Does the d-log and cinelike settings do anything for pictures? Im a photog and I shoot exclusively in raw....

I don't think so. Raw is independent of any camera style settings - sort of like the way raw works with a normal camera too.

 

[Sday]

I don't think so. Raw is independent of any camera style settings - sort of like the way raw works with a normal camera too.

I tested yesterday whilst shooting for a local fishing club and whilst shooting in raw doesn't seem to change with different custom sharpness, contrast and saturation levels but I did notice that changing d-cinelike and none made a huge difference. On none the saturation was increased and I tested on none with minus saturation figures, didn't change a thing.
From now on I will shoot photos in d-cinelike or trucolor

 

[Ziggy01]

I would like to get as many people involved in this discussion as possible as the internet is alive with these complaints and with our collective voices we can get this done and we all deserve better.

I know this is going to be hard for many of us as, like me you are probably so impressed with this little guy that a lot of us (like I said not everyone seems to be affected) are settling for a faulty product. And I get it. It is small, well built and is easier and more fun to fly than my p4. P4 Pro and Inspire 1, so how do you send something like that back? And if you’re like me, I ordered (and paid for it) on October 10th, but waited 50 plus days to get it (December 6th).

It also appears that some of the earlier shipments and video samples I saw (that originally sold me) seem… less than Phantom, but not by much. And as the old adage goes, “The best camera is always the one you have with you” I spend a lot of time in remote areas so I am ok with a little image sacrifice, but NOT what I am getting!

Ok so here it is and please chime in with observations and upload samples. The more proof we get the better chance we have of resolution.

A lot of us are having very serious focus and resolution issues with the Mavic. Regardless of how diligent we are on focusing (and YES we are taking the bubble off!) the image still seems very soft. So what I did was a close frame-by-frame analysis and in most cases each frame of the video looks more like a watercolor painting.

Here is a Dropbox link (Dropbox - Mavic Samples) to a folder that contains 4 short samples of video taken at 4096, 3840 and 1920 resolution. I also added the video named “Shot-Also-at-3840” that I originally shot that drove me to do a controlled test. Because in fairness the original subject matter of leafless trees is daunting detail for any camera much less this little guy and wanted to be fair. To that end my test consisted of something simpler. I did a hover in front of my house.

I have also provide representative frame grabs from each of the first 3 movies where you can clearly see the effect I am talking about. The 1920 video almost looks like a stylized posterize filter was put on in photoshop, and I believe that this is what is happening to all the videos just to a smaller extent.

PS. After sharing these samples with DJI that sent me an RMA within minutes and promised a new Mavic within 5 to 7 business days. I'll keep you all posted as to if this is lip-service or for real. Also if there is a moderator here or anyone that can get a DJI rep in this forum, PLEASE do!

PSS. So you know it is not me, here is another very capable guy on youtube showing the exact same "Water Color" issue.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

Hmm.

4K Test Footage from the DJI Mavic Pro is Gorgeous, NOT Blurry

 

[cardplayer71]

I am new with the Drone and video stuff, but I have noticed that in one of my videos that the image in some part of my video is good but like in the corner they look blurry.
should I contact DJI about this or there is an alternative solution. Ion the video I tried to focus a few times but still the image is not as expected.
Thank you for your help

Yes, this has been described before as a misaligned sensor. It is not aligned with the focal plane.

 

[newnan3]

I tested yesterday whilst shooting for a local fishing club and whilst shooting in raw doesn't seem to change with different custom sharpness, contrast and saturation levels but I did notice that changing d-cinelike and none made a huge difference. On none the saturation was increased and I tested on none with minus saturation figures, didn't change a thing.
From now on I will shoot photos in d-cinelike or trucolor

So youre saying that cinelike in raw reduced the saturation level for you?

 

[Sday]

Wasn't expecting it too but yes

 

[4wd]

It will look different on the monitor as you record but shouldn't affect the DNG.
That should only be changed by exposure really.

 

[Dale Rodgers]

Did a simple test in the living room tonight. Pointed the Mavic's camera at a lamp.

Filmed it in None (0, 0, 0) and turned over exposed (EV all the way to +3) and filmed. Put the video into my editor and viewed the scopes. Brightness was clamped at around 92%. Turned the EV all the way down (-3) and checked on scopes - clamped at 0% (a range of 92%).

Turned the contrast down to -3 (so now None 0, -3, 0) and repeated the above - over exposed clamped at around 88%, under exposed clamped at around 8% (a range of 80%).

Then used D-Log with contrast at 0 (0, 0, 0) and repeated the above - over exposed clamped at around 88%, under exposed clamped at around 7% (a range of 81%).

Then turned the contrast all the way down (so now D-Log 0, -3, 0) - over exposed clamped at around 82%, under exposed clamped at around 15% (a range of 67%).

Using D-Log narrows the dynamic range (I did however expect it to be worse). It may be using a different curve but you still can't get something for nothing from 8 bits of data.

The biggest throw away of dynamic range was lowering the contrast:

With the "None" profile it looks like you go from a range of 92% on my scopes to 80% when contrast is dialled down to -3. A loss of 12% that you'll never get back in post - its just been quantised away and possibly compressed away even further by the codec.

With the D-Log profile you go from a range of 81% on my scopes down to 67% with the contrast dialled right down. You'll never get this back in post.

My analysis may be over-simplistic and crude and doesn't take into account gamma curves and other fangled trickery. However, I believe you are already starting with less range for D-Log that you can never get back. Turning contrast down makes things even worse regardless of profile. Boosting in post is boosting information that is not there leading to quantisation effects (or colour banding).

With less range going into the codec then too many similar tones/shades may get compressed away causing mush. Then there's the noise reduction problems as well...

 

[BorisTheSpider]

Did a simple test in the living room tonight. Pointed the Mavic's camera at a lamp.

Filmed it in None (0, 0, 0) and turned over exposed (EV all the way to +3) and filmed. Put the video into my editor and viewed the scopes. Brightness was clamped at around 92%. Turned the EV all the way down (-3) and checked on scopes - clamped at 0% (a range of 92%).

Turned the contrast down to -3 (so now None 0, -3, 0) and repeated the above - over exposed clamped at around 88%, under exposed clamped at around 8% (a range of 80%).

Then used D-Log with contrast at 0 (0, 0, 0) and repeated the above - over exposed clamped at around 88%, under exposed clamped at around 7% (a range of 81%).

Then turned the contrast all the way down (so now D-Log 0, -3, 0) - over exposed clamped at around 82%, under exposed clamped at around 15% (a range of 67%).

Using D-Log narrows the dynamic range (I did however expect it to be worse). It may be using a different curve but you still can't get something for nothing from 8 bits of data.

The biggest throw away of dynamic range was lowering the contrast:

With the "None" profile it looks like you go from a range of 92% on my scopes to 80% when contrast is dialled down to -3. A loss of 12% that you'll never get back in post - its just been quantised away and possibly compressed away even further by the codec.

With the D-Log profile you go from a range of 81% on my scopes down to 67% with the contrast dialled right down. You'll never get this back in post.

My analysis may be over-simplistic and crude and doesn't take into account gamma curves and other fangled trickery. However, I believe you are already starting with less range for D-Log that you can never get back. Turning contrast down makes things even worse regardless of profile. Boosting in post is boosting information that is not there leading to quantisation effects (or colour banding).

With less range going into the codec then too many similar tones/shades may get compressed away causing mush. Then there's the noise reduction problems as well...

This is an interesting, if slightly crude test - thanks for doing it. Since being convinced to get off the d-log bandwagon of blindly going "hey, Beavis, it has a log profile, kewl" I'm actually satisfied with my results using cinelike to the point where I'm not sure I'm sufficiently motivated to do more testing using your method, but it seems like a somewhat scientific way to determine whether there's actually anything to be gained from a different profile. Perhaps I should even be shooting in none, so I might do the same with none vs cinelike I guess.