DJI Mavic, Air and Mini Drones
Friendly, Helpful & Knowledgeable Community
Join Us Now

Mavic Pro: why DJI doesn't fix this huge flickering/compression problem?

Men, probably you are right.
I made a test in a beautiful sunny day and the 4k at zero footage seems sharper at first sight but when you move around trees ecc...you can see much more compression and artifact than the same footage at 2.7k with sharpness at +1.
My conclusion now could be, as someone has already said, that at 4k the mavic increases the sharpness but without adding real details on the screen, just creating a "false" sharpness due to the algorithm inside....

I tried to upscale the 2.7 footage and the result is very good, not on par with the raw 4k but close enough.

I'll made some other tests and the situation is confusing me but, again, you are probably right and thank you because I'm learning many things.....in particular thank you to deadwings and fixaldo because I'm coming step by stpe to their conclusion ;)
 
Hi everyone.

I think maybe I have found a solution to the flickering issue. I use Davinci Resolve as an editor and using it's Temporal Noise Reduction feature I was able to completely erase the flickering from a 4K clip I tried it with, seemingly without affecting the look of the clip. I need to do more tests and also I'm going to read the manual so I can understand what all the different settings do but it was quite astonishing to render it out and see no flickering at all.

Hi, you really erase completely the flickering? I've tried and with temporal at 20 the situation is better but I can't erase it completely.....any other suggestion?
thank you
 
How do you think, is this ok filmed on Mavic Pro or I need to chenge some setting. Filmed on regular auto mode, but later put on DLog v.1 on FCPX
 
I've just reread all 9 pages of this post - phew!!

To the newbie this would still be very confusing as to the optimum settings.
The variables we are all talking about each directly affect the end result. Add one variable on top of another and the possibilities get mind-boggling.

We still don't have consensus on the best settings but our discussions & subjective tests on this forum is what makes the forum such a great tool. I'm trying to sum up mine & others testing conclusions so far, so have put it down here. Stop here if you expect a definitive conclusion at the end of this - I don't make one :)
I'm trying to get the theory correct before posting my conclusions.
...............................
Factors:

1. Brightness of day, cloud cover, subject colour & shadows (contrast) affects the dynamic range needed to record shadows & highlights. MP's sensor will only handle so much dynamic range, it isn't a high-end professional sensor. A sensor that will only record 8 stops of dynamic range will not record 12 stops of range not matter what you do. We struggle in Australia to get those lovely soft hues the northern Europeans seem to be able to capture - soft clouds are the free big diffuser in the sky. We get stuck with a huge big spotlight most of the time.

2. Choice of MP colour mode hugely affects how the "limited-dynamic-range 4.2.0-colour space" coming off the sensor is treated (via algorithm) before it is then compressed into the x.264 recording format (mp4/mov). Whilst Dji have given us colour mode tools so we can choose, I feel that most of these modes are designed for point & shoot users who just want a different look.

3. x.264 is excellent for giving apparent high quality visuals at low bit rates. It does this by throwing away masses of information. The x.264 GOP process in our MP workflow is what gives us our core flicker problem, on top of the limited bit rate compression that the MP encoding chip allows us to record in.

4. How you expose the footage has a huge influence on the end result. Is it best to lift your exposure to record detail in the shadows and throw away a bit of detail around the bright areas, or do the opposite? Is it best to expose evenly knowing darks and brights may not record everything. Or is it best to use dlog or cinelike hoping these algorithms will record as much dynamic range as possible and you'll be able to grade the picture back to normal. The small Go4 on-screen histogram is very hard to base exposure decisions on, and using dlog I appear to get top & bottom (within this histogram) areas of no image being recorded into. My background in film/digital producation tends to make me think the 4.2.0 colour space & x.264 hard coded recording bit-rates isn't capable of using these dlog/cinelike algorithms without compromising somewhere. When editing looking at the histograms of footage shot using dlog/cinelike it looks like I'm wasting about 25% of the recordable bandwidth. Top & bottom of histogram shows no information, meaning instead of recording colour information divided into 256 steps for each colour, I'm getting less than 200 - this increases the chance of banding. Others swear by dlog but for me the jury is still out.

5. Contrast, Colour values add another variable. Starting with a 4.2.0 colour space makes changing these values fraught with problems. Does reducing colour values in an already colour starved chain add to or reduce the original information coming off the sensor. Does reducing/increasing contrast make it harder to record dynamic range, and how is this affecting the colour profile selected. This to me is the last part in the puzzle to experiment with after other factors are worked out.

6. Sharpening appears to be added before(?) the colour space algorithm is applied, giving different effects to bright areas vs dark areas (water-colour-effect/overly sharp etc). Sharpening traditionally looks for adjoining pixels/areas contrast and enhances this contrast through quite clever algorithms, but can get ugly if too strong. If there is no discernible contrasty image in the dark areas you can't use contrast based sharpening. Tests I've seen here reducing camera sharpening, and then post sharpening have certainly reduced flickering but on all my different monitors made the image look very low resolution/quality. This has distracted my viewing experience way more than flickering.

7. Choice of post processing program (fcpx/neat/Davinci etc) and the various individual methods each have for sharpening/softening/temporal processing mean it's hard to compare each other's techniques on our own individual footage.

8. Youtube/Vimeo etc upload quality/compression-format and re-compression has a major impact on the quality/flicker. All our testing & subjective analysis should be done on pre-uploaded footage.

9. Our viewing monitors greatly affect how we perceive our tests. Looking at 4k/2.7k footage on a modern 55" TV is very different to looking at it on a HD/4k computer monitor. TV screens add masses of processing onto the footage in an attempt to make all the various qualities & resolutions fed to it via free to air, cable, internet etc look good and constant. Look at all the confusing names the manufactures use in their marketing materials discussing picture quality and processing "engines", every manufacturer has their own version. Computer screens can be quite brutal on video footage, making minor defects look major.
..............................................
Phew!! If you've got this far well done for perseverance. My post may be as complex to read as the previous 9 pages!!

My belief is DJI have no control over the image recording, they are locked into the sensor & video processing options by component manufacturers. It would be great if they would give us a little more technical information on their testing/reasoning behind the variables they have given us i.e. - why do we have cinelike & dlog - what shooting conditions are both algorithms optimized for etc.

Bottom line is we have an amazing little contraption capable of stunning results if all the variables line up. Just getting those variables right eludes us most of the time.
I am hugely grateful to all the various posters inputs & observations and certainly don't dispute anyones conclusions. We're all on the same side with trying to squeeze the best we can out of it.
Cheers.

Feel free to add to, correct or comment on any of the above.

:)
 
Last edited:
  • Like
Reactions: Pathogen
Thanks Oglo, a very excellent summary of things. Your mention of displays doing voodoo to our recordings is a point for people to consider with particular attention. It helps me understand why the DLOG + grading issue leading to banding is so rarely mentioned, when it is the reason I avoid using gamut crushing profiles as much as possible. It would appear that smart displays are applying a debanding routine.

The GOP shimmer/flicker artifact is aggravated by the fact that to get our exposure set correctly, we must use a very fast shutter speed. The high speed shutter setting records an extremely detailed series of frames, which is great for extracting still photos, but not so great for creating that smooth film like appearance of video.

We can use lens filters to allow lower shutter recordings, which can help, but cannot entirely solve the GOP artifacting problem.

If we record at 24fps, we end up with a GOP 'flash' every 8 frames, or 3 times a second, which can be quite noticeable depending on the content. However (yea I'm finally getting to it, I like to type, deal with it), what if we record at 48fps? Now we have tightened the GOP grouping to 6 flashes a second (I can smell your gray cells working already!), which is less noticeable unless you are a hummingbird. All that we need to do at this point is use our trusty video editor to blend the frames together to get our footage back to 24fps, and voila!

Even that clever method (some genius must have thought it up AMIRITE?) may not completely lower the GOP artifacting to an imperceivable level, or it may just not offer the right flavour of look. What then?

We can alternatively use post processing software (Windows in my case) to address the issue with exceptionally specific control. While there are many fine commercially available options, I prefer to roll my own custom solution using AviSynth. In any case the technique to master is motion blurring, otherwise know as 'temporal smoothing', which is to say, smoothing across time.

Combining the strategy of recording with a higher fps than you want as a final result, and also using some sort temporal smoothing process, the GOP shimmer artifact can be completely eliminated.
 
Last edited:
Pathogen, how we can record at 48fps if we are speaking mostly about 4k and 2k?
and maybe you have a specific knowledge of motion blurring (personally I wouldn't know where to start) :)
 
Gastone360 indeed it is the case only the 1080 and 720 modes allow for high frame rates. That combined with the GOP issue is why I feel DJI is dishonest to advertise the Mavic camera as 4k video capable.

Regarding motion blurring, the principle internal process involves taking two frames, and interpolating a new frame to go between them. That would double the frame rate artificially. Then two frames would be blended together, returning the frame rate back to normal. In reality though 24fps has to be converted internally to at least 384fps, and then selectively and repeatedly merged back to 24fps. That roughly creates the 24fps/50shutter effect which is a 50% on 50% off cinema look.

How one would apply that effect is dependant on the software you are using. Look for a 'motion blur' or 'temporal smooth' effect. I can't help much beyond that for commercial software, I use AviSynth to do it manually in a script.
 
Pathogen what a novel approach!
My basic understanding of temporal smoothing is that according to how you set your sliders/settings within this feature of an editing/processing program, the program looks at a series of frames either side of the current (central) frame and blends these back into the frame. It blends these frames in a diminishing scale according to how far away the frames are from the 'central' frame. You usually get to choose whether you have frames ahead as well as behind blended in. Obviously this smooths out detail/grain/noise and has a "streaky trail" affect on faster moving detail - a bit like motion blur but applied after the image is recorded at the post processing stage. It will work better on more static shots exhibiting GOP flicker & high ISO noise. It certainly will affect GOP 8 frame key frames, but using a temporal smoothing of at least 8 frame range is huge, giving a very smooth creamy feel to the image and reducing apparent resolution. It will work on some clips better than others.

Basic subjective tests by me yesterday certainly have me believing that 2.7k recording is giving me less flicker in the shadow areas. The bit rate per frame on 2.7k vs 4k calculations (done earlier in this post) mean the x.264 compression is favoring shooting in 2.7k.

A point aside from all this has popped up for me.
I had an extended 5 month period of no flying - a drowned MP being replaced plus holidays.
When I finally fired up the (brand new) replacement MP a couple of weeks ago and updated to latest firmware/software I was struck with noticing the images appeared (?!) to be cleaner/sharper/less-flicker.....
Either brain fatigue is setting in, or they are making un-announced incremental improvements somewhere in the chain. The most direct way to improve picture quality on the MP would be to up the bit rate, this isn't the case though. So possibly either my new sensor is higher quality, there is less vibration getting through to the sensor, or the colour profile algorithms & x.264 compression has been tweaked. This is a very, very subjective conclusion, definitely not based on hard proof. I would be keen to hear from owners of the new MPP who also have an original MP, as to whether they notice any visual difference. (This is is like letting the cat amongst the pigeons )
 
Oglo, regarding blurring out GOP artifacts, the method tends to vary depending on content, sometimes less is more, and at other times more isn't enough. :) In any case I stick with 1920x1080, because it offers the highest bitrate per pixel, and offers 48fps recording which makes motion based post processing generate more accurate results, and is less work for all post, and maybe even because I am a lucky owner of a Mavic with a slightly misaligned lens.

First I would like to ask, what revision of firmware is this new super duper Mavic running on? :)

Regarding temporal smoothing, please note that to get this to work without blurring everything horribly as you well described, we need to artificially increase the frame rate radically, via motion interpolation. I am finding that a minimum starting point is 192fps, with a 2 frame temporal smooth (which brings us down to 96fps), and then two stages of frame blending, which brings us back to 24fps. As you can observe here, the actual blur is only across 0.01 seconds. The remaining blending is done in a manner to exactly match what a slow shutter does. If the motion interpolation routine is failing due to rapid movement, we can still boost the number of frames it works on a fair amount before we suffer unwanted blurring.
 
I'm quite excited anche very very very happy!!!

I SOLVED the flickering problem at all with Davinci Resolve and temporal denoise. using temporal denoise at 15, once you set to 5 frames and "better" you can render the video and at the end you have a beautiful, clean, flicker-free image!!!!! Try it!!! From now I'mm shot in 4k and sharpness +1 without any problem

ps: I've changed my PC and now with a I7 7700HQ,16gb ram and a GTX1060 6gb I can render 1 minute in about 2 minutes and it is really fast
 
I'm quite excited anche very very very happy!!!

I SOLVED the flickering problem at all with Davinci Resolve and temporal denoise. using temporal denoise at 15, once you set to 5 frames and "better" you can render the video and at the end you have a beautiful, clean, flicker-free image!!!!! Try it!!! From now I'mm shot in 4k and sharpness +1 without any problem

ps: I've changed my PC and now with a I7 7700HQ,16gb ram and a GTX1060 6gb I can render 1 minute in about 2 minutes and it is really fast


Could you possibly upload somewhere a before/after full-res? Always interested in alternative solutions!
 
MAVIC PRO noise reduction - Google Drive

Temporal Denoise - 5 frames - better - values at 15 (with 20 is even better and more conservative but you start loose some details and sharpness)

The difference between the two is clear, so surely is a way.

But honestly to me the loss in quality is about the same if I was shooting directly with lower sharpness settings/resolution. Maybe depending on shooting one might be be slightly better than the other, but with the advantage of not having to wait all the time needed to process it.

I'll attach below a screen showing the differences after/before (looking at the far mountains, the grass and trees, details in the houses). The 4k still seems to me the usual artificial sharpening made by the Mavic to make videos 'pop' on small screens, but then if you need to basically 'undo' this sharpening process because it introduces artifacts (flicker), and doing so you basically introduce more artifacts (denoise), I think final image quality is potentially more degraded.

Anyway good to see one more possible solutions, good find! Personally I prefer to process the 'quality' of these files as little as I can considering the output from the Mavic is already lossy and not raw..
 

Attachments

  • Screen Shot 2017-12-05 at 00.49.54.png
    Screen Shot 2017-12-05 at 00.49.54.png
    4.3 MB · Views: 67
Thank you deadwing but I want explain that we all can shot in 4k 30fps but we need to use the temporal denoise to avoid the flickering.
It is the same video, one just from the sd card (before) and the other one rendered in Resolve with temporal denoise (after).

Probably you are right reading the footage from a scientific point of view but on a 4k screen I can see a clear difference, in general, between the 2.7 and 4k footage but this one was always ruined by this ugly flickering!!! And my thought was: never use 4k because is too noisy!!! :) ;:) Now I have the opportunity to use 4k without regrets! I repeat, on a 4k screen the difference in resolution is huge, even denoised...

We should make the same test in 2.7k and compare it with the 4k denoised.....
 
Last edited:
  • Like
Reactions: deadwing
Thank you deadwing but I want explain that we all can shot in 4k 30fps but we need to use the temporal denoise to avoid the flickering.
It is the same video, one just from the sd card (before) and the other one rendered in Resolve with temporal denoise (after).

Probably you are right reading the footage from a scientific point of view but on a 4k screen I can see a clear difference, in general, between the 2.7 and 4k footage but this one was always ruined by this ugly flickering!!! And my thought was: never use 4k because is too noisy!!! :) ;:) Now I have the opportunity to use 4k without regrets! I repeat, on a 4k screen the difference in resolution is huge, even denoised...

We should make the same test in 2.7k and compare it with the 4k denoised.....


Absolutely, that would be a final test. This period I'm busy in studio with production but if I'll get a chance I will.

If you will let us know, curious to see the final result in terms of resolution. To my findings is anyway possible to shoot 4k with -1 sharpness (or -2 for very complicated footage) without flickering, but somehow as soon as I set 4k on Mavic I see again this artificial sharpening. Is just too small lens to allow enough light in for 4k native, so I'm quite sure they do it upsizing with sharpening.

But testing will be indeed the way! :)
 
The best setting is 10 in temporal denoise because it limits flickering at minimum and you preserve resolution and sharpness...
 
Guys,
please take a look my post with what I believe is a same issue an let me know what do you think - it's the same issue or sth else? Link to video files and post is:
MavicPro Platinum - grain & poor video quality

I've done some colorgrading with LUT in Premiere Pro but would need your advice on hot to soften shrapening/flikcering in post like you do in Davinci Resolve.

The footage info is:

Grain 1.mp4
- latest firmware
- 4K, 30fps
- 1/320
- 100 ISO
- WB Cloudy
- D-log
- style settings 0,0,0

Grain 2, Grain3.mp4
- latest firmware
- 4K, 30fps
- 1/60
- 100 ISO
- WB Cloudy
- D-log
- style settings +1,-1,-1
- postprocessing color in PPP with LUT (only Grain 3.mp4)
 
Lycus Tech Mavic Air 3 Case

DJI Drone Deals

New Threads

Forum statistics

Threads
130,599
Messages
1,554,265
Members
159,605
Latest member
petravka