DJI Mavic Pro II

[sanveer]

18 hours ago, Robert Collins said:

The short answer is that both 4k FOV and 4k HQ shoot 10 bit with H.265.

Long answer.

I think Phillip Bloom confused matters by saying only 4k HQ shoots 10 bit. However, when you get your Mavic 2 Pro it asks you install a 23/08 firmware update. My guess is that this enabled FOV 10 bit but that Phillip Bloom was using earlier firmware (a couple of comments from DJI imply that a firmware update is needed.)

It is further complicated by the fact that nothing in the app (or online manual) says anything about 10 bit or 8 bit.

So I simply took some quick video in both 4k FOV and 4k HQ and looked at the underlying video exif using mediainfo. Mediainfo says both videos are 10 bit.

Could you share some link to this fact about the fov being the same for 8-bit and 10-bit. I could be wrong, but all documentation so far, proves things otherwise. 

[Robert Collins]

13 minutes ago, sanveer said:

Could you share some link to this fact about the fov being the same for 8-bit and 10-bit. I could be wrong, but all documentation so far, proves things otherwise. 

Sure.

Just to be clear what I said was that if you shoot DLog (which requires the h.265 format) both 4k FOV and 4k HQ record in 10 bit. It was dark and raining so I didnt fly the drone - just took some video. 01 is 4k HQ and 02 is 4k FOV (you can see the wider angle v the crop.)

And here are the original files to download (only about 400mb)

https://1drv.ms/f/s!Arx347dcSG470VzSNBiy1QaErAXn

The exif data shows both are 10 bit.

[sanveer]

3 hours ago, Robert Collins said:

Sure.

Just to be clear what I said was that if you shoot DLog (which requires the h.265 format) both 4k FOV and 4k HQ record in 10 bit. It was dark and raining so I didnt fly the drone - just took some video. 01 is 4k HQ and 02 is 4k FOV (you can see the wider angle v the crop.)

And here are the original files to download (only about 400mb)

https://1drv.ms/f/s!Arx347dcSG470VzSNBiy1QaErAXn

The exif data shows both are 10 bit.

The 1st file has a Crop wrt to the 2nd one. 

[robotfist]

On 8/29/2018 at 5:23 PM, Cliff Totten said:

One thing about noise reduction and detail - Ambarella noise reduction is not our friend. It's usually poor quality and ALL noise reduction destroys detail at certain levels. It's just a side affect that we need to live with. Noise reduction is also a DESTRUCTIVE process. Once you apply high amounts of noise reduction, it cannot be recovered or removed in post. Once noise reduction destroys detail, there is no plugin on planet Earth that can undue it when you get home. Therefore it's better to limit the amount of noise reduction in camera and capture more detail and noise in your file and then soften and apply noise reduction in post editing. Softening and image is the easiest thing to do in a video editor. Plus, you can surgically clean noise FAR more efficiently when editing than any camera can do live on the fly.

Applying +1 sharpening tells Ambarella's processor to ease up on this crappy noise reduction allowing more detail to come through. This gives you WAY more accurate options to soften and clean an that image in your editor.

Guys,...this one is going to shock you. I fell outa my chair on testing these. This are 400%+ crops on a chart print out.

Which one is the:

Mavic-II Zoom 28mm

Mavic-II Pro FOV Full

Mavic-II Pro HQ crop sensor mode

Phantom 4 Pro V2.0

All camera's set at 100 ISO (0db) and sharpness at +1

Wow......This is part of a shootout video I'm releasing next week.

So which is which? The very top image seems to have much less noise, so there is either mor aggressive NR happening there or that sensor is actually cleaner.  

[Moran]

On 9/1/2018 at 9:41 AM, Robert Collins said:

Sure.

Just to be clear what I said was that if you shoot DLog (which requires the h.265 format) both 4k FOV and 4k HQ record in 10 bit. It was dark and raining so I didnt fly the drone - just took some video. 01 is 4k HQ and 02 is 4k FOV (you can see the wider angle v the crop.)

And here are the original files to download (only about 400mb)

https://1drv.ms/f/s!Arx347dcSG470VzSNBiy1QaErAXn

The exif data shows both are 10 bit.

both files show 10bit information in exif because of the codec h265.

"Binning a 20mp output to 8mp takes a lot of processing power for real time. Add onto that having to output 10 bit, which is again resource intensive, and then again into a H. 265 format which is resource intensive, it would require a lot of processing power, which DJI probably realised ate a lot of battery power. The Inspire 2 can do this easily because it has 2 large 6S batteries powering it. A relatively small 4S battery - not so successful."
https://mavicpilots.com/threads/mavic-2-pro-4k-hq-vs-4k-fov.46068/#post-529138
 

the question is, does the camera records 10 bit colors on both HQ and FOV ? it's a matter of bandwidth. 
how can we check the actual colors recorded on footage ?  

 

[Luke Mason]

10 hours ago, Moran said:

both files show 10bit information in exif because of the codec h265.

"Binning a 20mp output to 8mp takes a lot of processing power for real time. Add onto that having to output 10 bit, which is again resource intensive, and then again into a H. 265 format which is resource intensive, it would require a lot of processing power, which DJI probably realised ate a lot of battery power. The Inspire 2 can do this easily because it has 2 large 6S batteries powering it. A relatively small 4S battery - not so successful."
https://mavicpilots.com/threads/mavic-2-pro-4k-hq-vs-4k-fov.46068/#post-529138
 

the question is, does the camera records 10 bit colors on both HQ and FOV ? it's a matter of bandwidth. 
how can we check the actual colors recorded on footage ?  

 

I used QCTools to check 10bit quantisation, I can confirm it's true 10bit colour.

[Grumble]

Tried to work out which of the @Cliff Totten images were from which camera/mode and gave up. But I will put them in order of what I think quality wise:

• Top one looks the best and sharpest on details, some moire
• Third one not as sharp but has the least moire
• Second one similar sharpness to the third but more moire
• Last one appears to have the most blocking, especially on the large dark area, similar sharpness and moire to top and second

Looking forward to seeing the actual results to see if it changes my purchasing. I also assume this is just for video quality and stills will differ?

[Robert Collins]

2 hours ago, Grumble said:

Tried to work out which of the @Cliff Totten images were from which camera/mode and gave up. But I will put them in order of what I think quality wise:

• Top one looks the best and sharpest on details, some moire

Dont agree at all. The first one is massively 'oversharpened'. You can tell because 'sharpening' increases contrast at the 'edges' within an image. Oversharpening results in halos and the first image has halos in spades (or in the case of the playing card - diamonds.) This sort of sharpening results in horrible motion artifacts.

[webrunner5]

#3 must be the 1" one. Seems to have the most resolution?

[Attila Bakos]

12 hours ago, Luke Mason said:

I used QCTools to check 10bit quantisation, I can confirm it's true 10bit colour.

Thanks for mentioning QCTools, I didn't know about it and it seems to be an awesome tool for analyzing video.

[Grumble]

11 hours ago, Robert Collins said:

Dont agree at all. The first one is massively 'oversharpened'. You can tell because 'sharpening' increases contrast at the 'edges' within an image. Oversharpening results in halos and the first image has halos in spades (or in the case of the playing card - diamonds.) This sort of sharpening results in horrible motion artifacts.

Fair enough, I find them all a bit hard to judge accurately if I'm honest. Things might look different between them all at 100% rather than 400% crops as well. Roll on the final results!

[Emanuel]

 

 

...

And here's one of my favs of this baby to rivalry with Pocket 4K the title of 'product of the year' IMHO:

Two tips for the best outcome to extract from these birds...   

As @Robert Collins has wisely hinted up there, go soft blurring your footage @post towards the right balance as much as necessary (use PotPlayer + Ctrl / R to preview the whole thing for this last sample as you wish) and don't throw away the chance coming already from the previous first generation going along with tap to focus to pair and mitigate the issue;

from minute 3:45 on:

(you'll see how non-sense the sharpness mantra can be; strictly under a very meaningful cinematic perspectve when it is where the unexpurgated frugal concept of beauty itself rises and stands -- funny and pretty ignorant times we're living in : P)

 

[Robert Collins]

1 hour ago, Emanuel said:

 

 

The image quality in the first 20 second intro to this video looks really good to me. You can see some great dynamic range from a 1 inch sensor and really good color tonality. This was DLog, 4k HQ and sharpness set to '0'.

[deezid]

10 hours ago, Robert Collins said:

The image quality in the first 20 second intro to this video looks really good to me. You can see some great dynamic range from a 1 inch sensor and really good color tonality. This was DLog, 4k HQ and sharpness set to '0'.

Has somebody seen any footage with sharpening set to something below 0?
On the P4P I have to set sharpening to -2 and codec to H264 to have the least amount of sharpening without blurring detail. At -3 fine detail and textures in shadows are messed up (like the Mavic Pro does) and an overall blur of low contrast textures when set to H265.

Maybe DJI fixed these issues...

[wolf33d]

 

Who said not sharp? As sharp as P4 pro. Image is great. 

[deezid]

7 hours ago, wolf33d said:

 

Who said not sharp? As sharp as P4 pro. Image is great. 

Well it is less sharpened technically, but also has a lot less noise reduction!
So you're actually seeing more detail especially in shadow areas.

[ErikSwan]

With a little lot of temporal noise reduction, I'm getting some pretty impressive results out of this sensor.

Granted, there are some artifacts from the heavy NR, but it still looks quite usable to me.

Shot 30 minutes after sunset. For an idea of how little light there was, look at the headlights on the hill in the background at the start of the clip, the various phone screens you can see throughout, and the city lights in the background towards the end.

I haven't owned any of DJI's previous drones but I have a hard time believing any of the 1/2.3" sensors would have been able to get anything remotely looking like this.

 

 

[sanveer]

I realised that the Mavic 2 Pro saves HDR photos with14 stops of dynamic range, as JPEGs . If the Mavic saved these as 10-bit HEIFs instead, it could improve things for grade, considerably. It could be the perfect middle ground between JPEG and RAW, with boosted dynamic range. 

[Andrew Reid]

Don't know if this has been mentioned before, but I wish I didn't have to buy two separate drones just to get the two different camera heads.

[sanveer]

7 minutes ago, Andrew Reid said:

Don't know if this has been mentioned before, but I wish I didn't have to buy two separate drones just to get the two different camera heads.

That and making the cakera swapping seem so company process confined, is a ice way yo make mote profits.