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?

Hi, simply put temporal denoise at 10 (best) in Davinci Resolve Studio ad a first node and on your 4k +1 sharpness footage and save it in 4k max resolution. In this way you'll have clean footage to work with and you can do the rest you need

I always clean my footage now with Davinci and the results are better and way faster than Neat Video (tried personally both)
 
You mean that you put the chroma and luma levels @10? You keep motion at 0?
Because it depends strongly on the footage.


Hi, simply put temporal denoise at 10 (best) in Davinci Resolve Studio ad a first node and on your 4k +1 sharpness footage and save it in 4k max resolution. In this way you'll have clean footage to work with and you can do the rest you need

I always clean my footage now with Davinci and the results are better and way faster than Neat Video (tried personally both)
 
You mean that you put the chroma and luma levels @10? You keep motion at 0?
Because it depends strongly on the footage.

My settings are:
Temporal denoise
Frames 5
Est. Type better
Motion range medium

Threshold
Luma 8 (it is ok between 8-15)
Chroma 8 (the same as Luma)
Motion – not touched
Blend not touched

Spatial NR – not touched

MAVIC PRO - Google Drive

I put a comparison between 2.7k not touched and 4k denoised with theese settings above
Luma and Chroma is basically on personal tastes but I think between 8-12 is the right way to preserve details for 4k shoots.
Unfortunately movement in this footage is very slow but you could see the differences.....
bye
 
  • Like
Reactions: alexsolich
My settings are:
Temporal denoise
Frames 5
Est. Type better
Motion range medium

Threshold
Luma 8 (it is ok between 8-15)
Chroma 8 (the same as Luma)
Motion – not touched
Blend not touched

Spatial NR – not touched

MAVIC PRO - Google Drive

I put a comparison between 2.7k not touched and 4k denoised with theese settings above
Luma and Chroma is basically on personal tastes but I think between 8-12 is the right way to preserve details for 4k shoots.
Unfortunately movement in this footage is very slow but you could see the differences.....
bye
Thank you so much!
 
My settings are:
Temporal denoise
Frames 5
Est. Type better
Motion range medium

Threshold
Luma 8 (it is ok between 8-15)
Chroma 8 (the same as Luma)
Motion – not touched
Blend not touched

Spatial NR – not touched

MAVIC PRO - Google Drive

I put a comparison between 2.7k not touched and 4k denoised with theese settings above
Luma and Chroma is basically on personal tastes but I think between 8-12 is the right way to preserve details for 4k shoots.
Unfortunately movement in this footage is very slow but you could see the differences.....
bye
And to sum it up your settings on the mavic side 4k @24fps +1-0-0 d-log?
I personally tried neat image calibrated and davinci but without luck yet
My mavic settings were ranging from 2,7k 0-1 sharpness ranve -3-0 contrast and range -3-0 for saturation , i tried the same for 4k . I always expose ETTR as I do for micro 4/3, and always correct in post.
Did not go below 0 sharpness . Even at 0 i see a lot of compression artifact. Yet I screw up something it seems .
I can send you a vid to evaluate.
Best regards
Alex
 
What I already see, that grading flat 8 bit footage is a PITA with mavic like quality, you might actually loose some color data because one is unable to get back the right colors without compromising image quality and introding flickering. But it's true that using 2.7k makes grading workflow (even aggressive) more pleasing.
Although there is a slight difference between 2.7 and 4k but you loose the edge of 4k when you try to edit the colors aggressively . So stick with higher res and accept that you can do narrow adjuatments in post or have it in 2.7k have greater grading possibilities
For me the slight difference in resolution with greater restrictions in post aren't worth it . Your viewers will be more pleased with a finely graded non flickering video, i guess. Try not to satisfy the random pixelpeeper on vimeo.

Cheers
 
If you guys interested I will run a comparison between davinci resolve denoise and fcpx neat video plugin on iMac Pro (10core 64gb, Vega64 16gb) just to see the difference, same 4k footage. And comparing with the same wf grading etc. with 2.7k footage. Reply if interested.
Cheers

Alex
 
Hey Guys, if you are interested here are some dlog footage in 2.7 and 4k with various sharpness settings contrast and sat @ -3.
Feel free to edit and do whatever you want.
Enjoy

mavic - Google Drive
 
Hey Guys, if you are interested here are some dlog footage in 2.7 and 4k with various sharpness settings contrast and sat @ -3.
Feel free to edit and do whatever you want.
Enjoy

mavic - Google Drive


Awesome, thanks! I have been doing something similar. What were your conclusions between resolve and neat video? I do not have neat video to compare to but would get it if it is noticeably better.

I am still going back and forth between 2.7K and 4K. On my iMac 5K, I can clearly see the resolution difference but also see the increased flickering/compression. With resolve, I think I can get the flickering down on the 4k@30fps to where I don't see it in a 4k window and I *think* it still has more detail. Still working on it.

Really big difference between 4k@30 and 2.7k@ 24 from what I see. From the math I did, the compression ratio for 4k@30fps is 33.3 and it is only 17.8 for 2.7k@24 fps. (4k@24fps is about 26.6 and 2.7k@30fps is about 22.2) ... if I did my math right!.
 
Awesome, thanks! I have been doing something similar. What were your conclusions between resolve and neat video? I do not have neat video to compare to but would get it if it is noticeably better.

I am still going back and forth between 2.7K and 4K. On my iMac 5K, I can clearly see the resolution difference but also see the increased flickering/compression. With resolve, I think I can get the flickering down on the 4k@30fps to where I don't see it in a 4k window and I *think* it still has more detail. Still working on it.

Really big difference between 4k@30 and 2.7k@ 24 from what I see. From the math I did, the compression ratio for 4k@30fps is 33.3 and it is only 17.8 for 2.7k@24 fps. (4k@24fps is about 26.6 and 2.7k@30fps is about 22.2) ... if I did my math right!.
2.7k bitrate is lower than 4k.. Only 40 mbps. 4k bitrate is 60mbps, both cbr. The flicker is a common issue of x264 codec. With the lastest fw, the. 200, and the 100, you have to set sharpness to 0 or - 1, ando the flicker effect dissapears.
 
2.7k bitrate is lower than 4k.. Only 40 mbps. 4k bitrate is 60mbps, both cbr. The flicker is a common issue of x264 codec. With the lastest fw, the. 200, and the 100, you have to set sharpness to 0 or - 1, ando the flicker effect dissapears.

Thanks Sebastian, I understand the bitrates, that is how I calculated the compression ratios. (Note: I think the rate for 2.7k is 45mbps ....)

e.g. for 4k@30fps
3839*2160=4.31Mpixels/frame
4.31Mpixels/frame*30 frames/sec = 249Mpixels/sec
249Mpixels/sec * 8 bits/pixel = 1995 Mbits/sec (unencoded rate)
(1995Mbits/sec) / (60Mbits/sec) = 33.2

For me, with the sharpness at 0 or -1, I usually get very soft blob type effects in the shadows. As others on this thread have observed, it seems like the Mavic applies noise reduction whenever the sharpness is <1. So far, I am having better luck trying to knock down the flicker but maybe I should investigate 0 more. Have you noticed any difference between 200 and 100 regarding sharpness / noise reduction.
 
Thanks Sebastian, I understand the bitrates, that is how I calculated the compression ratios. (Note: I think the rate for 2.7k is 45mbps ....)

e.g. for 4k@30fps
3839*2160=4.31Mpixels/frame
4.31Mpixels/frame*30 frames/sec = 249Mpixels/sec
249Mpixels/sec * 8 bits/pixel = 1995 Mbits/sec (unencoded rate)
(1995Mbits/sec) / (60Mbits/sec) = 33.2

For me, with the sharpness at 0 or -1, I usually get very soft blob type effects in the shadows. As others on this thread have observed, it seems like the Mavic applies noise reduction whenever the sharpness is <1. So far, I am having better luck trying to knock down the flicker but maybe I should investigate 0 more. Have you noticed any difference between 200 and 100 regarding sharpness / noise reduction.
100 and 200 no difference. GO. 4.1.22 vs 4.1.20 and 4.1.18 are completely different.

4.1.22 has the same sharpness behavior than 4.1.10, and older.
This version should be an evolution of that.
 
100 and 200 no difference. GO. 4.1.22 vs 4.1.20 and 4.1.18 are completely different.

4.1.22 has the same sharpness behavior than 4.1.10, and older.
This version should be an evolution of that.


Thanks, I am still on 4.1.15. How would 4.1.22 compare to that? I have hesitated upgrading since mine is working so well but would consider it if the video quality has advantages.
 
Thanks, I am still on 4.1.15. How would 4.1.22 compare to that? I have hesitated upgrading since mine is working so well but would consider it if the video quality has advantages.
4.1.15 : 0, is 4.1.22 +1
Noise filter is disabled at +2 now.
To record landscapes and high altirude, you should use +1 to avoid noise.
To record close objects, 0 is the Best setting.
If you use neat video, +1 is the best.
Trees and grass are a true nightmare to neat video.


Sorry for my english..
 
It means very simply that if you shoot at 30fps (for example), you ideally want shutter speed at 60, and these settings are all adjustable in the video settings (just under the recording button on the DJI app on iOS).

Is not essential with all that compression going on, but surely is the icing on the cake :)
Can you help me. The video is zero. I do not know anything about photograpy.
Give me the standard settings so that I can get video to see.
thanks
 
My settings are most of the time true color with sharpness +1 and zero the rest. I usually don't shoot on dlog but I'm starting in these days
I tried both neat video and DaVinci but, at iso 100, DaVinci is more detailed and more natural and...... Way way way faster

P. S I always use polar pro nd filters
Pps I'm interested in the comparison between DaVinci and neat video in denoising.. Because I prefer DaVinci a lot
 
Last edited:
My settings are most of the time true color with sharpness +1 and zero the rest. I usually don't shoot on dlog but I'm starting in these days
I tried both neat video and DaVinci but, at iso 100, DaVinci is more detailed and more natural and...... Way way way faster

Even if you have a high end machine, i would restrict the usage of neat video for only the most critical footage. Or spend a night or two rendering out a clean footage in fcpx and then edit in davinci. Actually my workflow is the latter now. Apply only the neat video for the ungraded video and then apply all the grading and fx in davinci.

Cheers
 
yes, for this reason I use Davinci, because in a few minutes it cleans the footage and then I can work with clean picture in post. I always clean the footage in section for this reason
 
Hello everyone, after doing some even more 'scientific' tests I've got some results.

They match with others on this forum, but they definitely disagree with the great video we all saw suggesting that sharpness +1 is the 'natural' (no processing) setting.

First of all the tests I'm doing, like in my first post video, are to put encoder to the extreme situation: contrasty scene, moving nearby small fine details (greenery moved by wind) and on same imagine a more far (but not too much) equally detailed background (loads of trees) in the shadows nearby a lit sky. In this demanding situation, using sharpness +1 underlines the work behind the compression and the GOP frames are the 'sharpened' ones while the others are very compressed, and that leads to all 'flickering' problems (details, edges, shadows/lights) we see everywhere on this forum (and that I found more disappointing than I expected).

To be honest all the nice Youtube footage I've seen, I realise now, is filmed from very high in the sky, in stable light, and so you get a frame without much fine detail, without moving lights, and all frame moving more uniformly and slowly (not bottom half moving 'fast'), or other are test shots on a bunch of trees or 'static' scenes with drone not moving. Now that I know that defect, I spotted the same 'shadows/light' and details flickering in loads of these, and yesterday I even spotted it clearly on a recent BBC documentary footage that was filmed by a drone!

I think the problem here is that the small sensor cannot (understandably) get too much detail, and DJI try with sharpness to save situation a bit, tuning it with the classic shots everyone will do (high in the sky), but it shows the compression limits on any more 'demanding' situation.

Someone else (many) here said that the ideal settings for sharpness was -2, and I have to say it's actually true. I found that sharpness -2 produces the most 'natural' result, considering that is much better to actually try and sharpen a bit in production, where you have much more control and you can actually sharpen more uniformely ALL frames not just the key ones (and so no flickering).

I tried all sharpness settings in both 4k and 2k at 30fps, and interesting is also that 0 in 2k has same 'defects' of -1 in 4k, -1 in 2k has same 'defects' of -2 in 4k and so on, really almost identical, but once you arrive at -3 on 4k (and so my ideal -2 in 2k) you get a much worst result in 4k than in 2k. Someone on this forum said -2 -1 -1 2k 30fps is the ticket, indeed I found the same. I've done all tests on the same field and same flight path. I'll probably use 2k -1 sharpness when doing less demanding footage in terms of details/movement/light, but the sweet spot is for me -2.

To be honest working on this I started realising that anything sharpness 0 and up looks to me a lot 'digitally' processed..it has that 'wow' factor at first but looks more 'cool' than natural/film, is a lot like all those HDR pictures...is cool and has wow factor to the unaided eye, but it looks very unnatural and 'fake' to me, as I like more 'real' photo or video.

Below the same place of my first video filmed in 2k -2 -1 -1 30fps, conformed 24fps and graded and sharpened a bit to be still natural and 'film' like (to my taste), the sky is blown out as I was filming in Natural (I'll be back in DLOG as it's not the problem and has a wider dynamic), but in the end I wanted sky very bright to be in contrast with the critical shadow area of the trees:

Well thats a turn up! vast improvement on river shots Almost broadcast quality, which is impressive from such a tiny sensor! Good and useful analysis. will check it out also.
 
  • Like
Reactions: Gray_seagull
Apologies if this has been posted already. Reasonably certain the fix for this is first, a firmware refresh (download and use Assistant 2) followed by IMU recalibration, then compass recalibration then lastly a gimbal recalibration. Should also return to factory settings for camera if the refresh does not do that automatically.
Just went thru this with my unit.
Had problems mid 2017 with similar video. After a trip to Costa Rica, I was dismayed to find most of my great footage was bound for the trash. I had submerged the drone in a nearby lake last summer but dried it out and it seemed to function fine so I attributed the problem to that. Used my last DJI Care to get a replacement unit only to find it did the same thing. Panic.
Followed the calibration above and it fixed her all up.
Seems some if not most of the firmware updates MUST be followed by this process to get the unit to function properly again.
 
Lycus Tech Mavic Air 3 Case

DJI Drone Deals

Forum statistics

Threads
130,931
Messages
1,558,041
Members
159,937
Latest member
potenthockey0