Please select Into the mobile phone version | Continue to access the computer ver.
Unacceptable Image noise at night.
2344 20 2018-8-31
Uploading and Loding Picture ...(0/1)
o(^-^)o
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
Offline

So, I held off for a long time to buy the Mavic 2 Pro, after feeling somewhat disappointed with the original Mavic. That being said, my Mavic 1 was actually amazing at taking 'nightlapses' and I swore by the fact that was and is the best drone for such a job.

I ordered the M2P with high hopes and unfortunately I've hit an issue which I didn't expect; image noise.

I have now, in the 48 hours of owning this M2P, taken thousands of stills at night, using POI and hyperlapse and consistently see noise in the images, which is then seen in the final video. I understand that JPG noise is accepted, but as a photographer this isn't normal whatsover. These are bright white spots in the middle of black areas and clear bands are visible, almost like a pattern.

The M2P was then put in a pitch black room and I took a JPG+RAW image. The Raw image came out as expected, with no noise whatsoever. The JPG looked like a starrey night once I started to zoom in.

To say I'm disapponted is an understatement. The original Mavic at 12mp with the cmos sensor, delivers better long exposure JPG stills than the brand new M2P...

The Phantom 4 Pro does not have this JPG noise problem. I was expecting a similar, if not better performance from the MP2. I've now added a P4P photo with the same settings/scenario.

I can't see this being a firmware fix. Images attached.

Photo - f2.8, ISO 100, 1.3"

Image attached.
Original Photo - https://we.tl/t-rNCjeDHCNG

M2P Photo taken in dark room (f2.8 ISO100 2")  (looks nice and black, but zoom in!) - https://we.tl/t-WduWVrcPCj

P4P taken in dark Room (f2.8 ISO100 2") - https://we.tl/t-euPYfsfIgK (totally acceptable colour of noise)

Edit: I have inverted the the above black photo, (see attatched) adjusted the levels to highlight the noise and there's a clear pattern at the edges of the image. To replicate, use Photoshop>Invert>Levels>Input Levels 0 | 0.01 | 255



Screenshot 2018-09-01 at 00.16.13.png
DJI_0239.jpg
2018-8-31
Use props
msavic6 5
lvl.4
Flight distance : 278583 ft
Canada
Offline

If you don’t notice the issue in the RAW images then it’s a by product of the image processing done before it is converted to a JPEG. This means the JPEG processing  can be modified with a firmware update. Regardless of this issue, the output Hyperlapse from the Mavic 2 Pro is only 1080p so it doesn’t fit my needs anyway. I am happy that it gives the option to output the RAW images which I can then post process myself and assemble into a hyperlapse in a much higher resolution.
2018-8-31
Use props
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
Offline

msavic6 5 Posted at 2018-8-31 16:05
If you don’t notice the issue in the RAW images then it’s a by product of the image processing done before it is converted to a JPEG. This means the JPEG processing  can be modified with a firmware update. Regardless of this issue, the output Hyperlapse from the Mavic 2 Pro is only 1080p so it doesn’t fit my needs anyway. I am happy that it gives the option to output the RAW images which I can then post process myself and assemble into a hyperlapse in a much higher resolution.

I can't work out what point you're making, especially as you're assuming that I'm using a 1080 output of a hyperlapse. I am talking about the JPGs, which are very noisy.
2018-8-31
Use props
No Man's Drone
Second Officer
Flight distance : 301545 ft
United States
Offline

CloudVisual Posted at 2018-8-31 16:19
I can't work out what point you're making, especially as you're assuming that I'm using a 1080 output of a hyperlapse. I am talking about the JPGs, which are very noisy.

I think he meant to say that the Hyperlapse video when stiched automatically by the DJI Go App is just 1080p. Of course, you have the option to save the original images and stitch them yourself for a higher resolution output.
2018-8-31
Use props
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
Offline

No Man's Drone Posted at 2018-8-31 16:46
I think he meant to say that the Hyperlapse video when stiched automatically by the DJI Go App is just 1080p. Of course, you have the option to save the original images and stitch them yourself for a higher resolution output.

Yeah, that processed video is no use to me..! I'm talking about the images and truthfully I'm actually starting to think that this is a big problem as the pattern in the noise looks to follow the shape of the lens...
2018-8-31
Use props
msavic6 5
lvl.4
Flight distance : 278583 ft
Offline

CloudVisual Posted at 2018-8-31 16:19
I can't work out what point you're making, especially as you're assuming that I'm using a 1080 output of a hyperlapse. I am talking about the JPGs, which are very noisy.

Are you not using the RAW images from the hyperlapse? I have my M2P set to save the Original RAW images.
2018-8-31
Use props
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
Offline

msavic6 5 Posted at 2018-8-31 16:55
Are you not using the RAW images from the hyperlapse? I have my M2P set to save the Original RAW images.

I'm saving the original JPGs. I'm not shooting in RAW.

Are you sure you mean RAW? I'd be surprised if you're shooting a TL in RAW, given the 5 second delay
2018-8-31
Use props
msavic6 5
lvl.4
Flight distance : 278583 ft
Canada
Offline

CloudVisual Posted at 2018-8-31 16:59
I'm saving the original JPGs. I'm not shooting in RAW.

Are you sure you mean RAW? I'd be surprised if you're shooting a TL in RAW, given the 5 second delay

Yes I mean RAW, I did a test hyperlapse yesterday which resulted in a 5 or so second clip and 125 Frames taken, took roughly 4.5 minutes to complete. I have a folder with 125 RAW (DNG) Photos each of size 40mb. The only downside is the storage required but I can’t imagine not using the RAW images as I’ve never been a fan of DJI’s JPEG processing.
2018-8-31
Use props
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
Offline

msavic6 5 Posted at 2018-8-31 17:09
Yes I mean RAW, I did a test hyperlapse yesterday which resulted in a 5 or so second clip and 125 Frames taken, took roughly 4.5 minutes to complete. I have a folder with 125 RAW (DNG) Photos each of size 40mb. The only downside is the storage required but I can’t imagine not using the RAW images as I’ve never been a fan of DJI’s JPEG processing.

There's no way I'm going to be shooting a TL in RAW, as the gap between shots is too great and the amount of shots are too few to make anything worthwhile. I need a minimum of 10 seconds, which is impossible with the amount of time a drone can stay in the air with 5 seconds between shots.

A JPG timelapse is no different in terms of bitrate to a ProRes video, which is above and beyond the requirements of almost any high quality video. What I am trying to say is that there is unnecessary noise within the JPG files, noise which shouldn't exist at all. Bright white spots in a 100% black image.

Please download the two black images I posted above. One is the M2P and the other a P4P. Zoom all the way in and compare the noise. The P4P has the odd group of black pixels which are out of place. The M2P looks like a galaxy. The two drones have nearly the exact same sensor, lens and processing.
2018-8-31
Use props
msavic6 5
lvl.4
Flight distance : 278583 ft
Canada
Offline

CloudVisual Posted at 2018-8-31 17:18
There's no way I'm going to be shooting a TL in RAW, as the gap between shots is too great and the amount of shots are too few to make anything worthwhile. I need a minimum of 10 seconds, which is impossible with the amount of time a drone can stay in the air with 5 seconds between shots.

A JPG timelapse is no different in terms of bitrate to a ProRes video, which is above and beyond the requirements of almost any high quality video. What I am trying to say is that there is unnecessary noise within the JPG files, noise which shouldn't exist at all. Bright white spots in a 100% black image.

I don't understand your reasoning for not wanting to shoot in RAW. My workflow consists of importing the RAW images into Lightroom, processing them to my taste, batch exporting the files as JPEGs and then creating the actual time-lapse in Premiere. The only legitimate downside is the storage space required and the additional time needed to export from Lightroom.

Regarding your comparison shots, I do see the white specs of noise in the image when zooming in.  Since you mentioned it only appears in the JPEG shots, I am 100% convinced this not an issue with the LENS itself or the actual sensor itself but with the JPEG engine being utilized and in turn the noise reduction. This is something that they can fix with a firmware update or at the very least tweak significantly. Until they release an update to fix this, I think your only workaround for night time-lapses is to shoot in RAW and do the noise reduction yourself.
2018-8-31
Use props
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
Offline

msavic6 5 Posted at 2018-8-31 17:31
I don't understand your reasoning for not wanting to shoot in RAW. My workflow consists of importing the RAW images into Lightroom, processing them to my taste, batch exporting the files as JPEGs and then creating the actual time-lapse in Premiere. The only legitimate downside is the storage space required and the additional time needed to export from Lightroom.

Regarding your comparison shots, I do see the white specs of noise in the image when zooming in.  Since you mentioned it only appears in the JPEG shots, I am 100% convinced this not an issue with the LENS itself or the actual sensor itself but with the JPEG engine being utilized and in turn the noise reduction. This is something that they can fix with a firmware update or at the very least tweak significantly. Until they release an update to fix this, I think your only workaround for night time-lapses is to shoot in RAW and do the noise reduction yourself.

I mentioned above: There isn't enough flight time at one photo every five seconds to make a timelapse which is greater than ten seconds. Furthermore, ProRes-level bitrate is more than enough.

I can't stress this enough, but the JPG photos should not have the white pixels. They have a pattern which is consistent with the lens and form a halo and can be seen as moiré in a dark video.

You don't need to tell me about the uses of RAW, I've worked as a photographer for years - I swear by RAW! However, for video, compiled JPG stills are more than enough, but in this instance, there is a serious issue with the JPGs from this camera.
2018-8-31
Use props
msavic6 5
lvl.4
Flight distance : 278583 ft
Canada
Offline

CloudVisual Posted at 2018-8-31 17:48
I mentioned above: There isn't enough flight time at one photo every five seconds to make a timelapse which is greater than ten seconds. Furthermore, ProRes-level bitrate is more than enough.

I can't stress this enough, but the JPG photos should not have the white pixels. They have a pattern which is consistent with the lens and form a halo and can be seen as moiré in a dark video.

Have you noticed the odd white pixels in any other shots that weren't at night? I've looked through the shots I've taken so far and don't notice it in any of the JPEG's I have.
2018-8-31
Use props
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
United Kingdom
Offline

DJI - Here's the inverted M2P black image with clear patterns at the edges. I have repeated this with a P4P JPG and there is almost no noise from a black photo.

2018-9-1
Use props
GDL
First Officer
Flight distance : 2768694 ft
Hong Kong
Offline

How about day time jpg photo? Is any noise on there?
2018-9-1
Use props
gnirtS
Captain
Flight distance : 5712575 ft
United Kingdom
Offline

I can put any of my DSLRs in a completely dark room and expect to see noise on the image regardless of ISO. 100% black is always going to get thermal noise on any sensor.  Thats not a fair test.
That's why NR is *always* used in the workflow.
Thats why nobody actually tests that because it shows nothing.  There are specific scenes and charts for measuring noise but "fully black" isn't one of them.

The two drones have nearly the exact same sensor, lens and processing.


Actually, they have completely different sensors, completely different lenses and completely different image processing SOCs.
2018-9-1
Use props
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
Offline

gnirtS Posted at 2018-9-1 02:40
I can put any of my DSLRs in a completely dark room and expect to see noise on the image regardless of ISO. 100% black is always going to get thermal noise on any sensor.  Thats not a fair test.
That's why NR is *always* used in the workflow.
Thats why nobody actually tests that because it shows nothing.  There are specific scenes and charts for measuring noise but "fully black" isn't one of them.

I'd like to point out that I said 'Unacceptable amount of Noise'

I want to know why the noise is white, whereas the noise present on a P4P is black/dark and in line with the colour of the shot. There is no reason why the noise should be speckled white in a dark shot.

I've just tried my 5D3 in the same environment and yes, I totally agree about some noise being present as it's a JPG and that is the nature of this format. HOWEVER the noise isn't white and making the image look like a clear night sky.
2018-9-1
Use props
gnirtS
Captain
Flight distance : 5712575 ft
United Kingdom
Offline

Just image processing choice.  Depends how it wants to handle random thermal noise.

JPG noise doesnt bother me at all (once you start putting noise through lossy compression its going to look awful) and who spending money on a M2P with a good camera will use jpg anyway?

Again id like to see real life images (which arent 100% black) to test.
2018-9-1
Use props
CloudVisual
Captain
Flight distance : 97545420 ft
  • >>>
Offline

gnirtS Posted at 2018-9-1 03:07
Just image processing choice.  Depends how it wants to handle random thermal noise.

JPG noise doesnt bother me at all (once you start putting noise through lossy compression its going to look awful) and who spending money on a M2P with a good camera will use jpg anyway?

Just for clarity, I use JPG for timelapses only. I'm not shooting them in RAW as I don't want to have a five second delay between shots and only a maximum of 10 seconds for the export.

Shooting a JPG timelapse is the same bitrate as ProRes and I can assure you that this looks nothing like ProRes-level quality.

These videos aren't going to YouTube or somewhere lossy, they're shot for stock footage and most agencies will spot the noise and refuse this quality of video.
2018-9-1
Use props
paul2660
Second Officer
Flight distance : 10331 ft
  • >>>
Offline

If noise is white odds are because it's a Sony sensor.  Sony is notorious for white dot issues in the modern sensors on longer exposures.

See Nikon D810, Pentax K1, Sony A7rII and A7rIII.  

If you see the white dots in raw, it's a form of noise that can be removed with Capture One/Phase One, with single pixel noise reduction.  Only tool I have found that removes such noise without loss of details.  If the white dots in the jpgs, they are fixed due to jpg format and nothing will easily remove them.  

They will get worse as sensor heats up.  

Paul C
2018-9-1
Use props
DJI Wanda
Administrator
Offline

Hi We apologize for the inconvenience. Did you try to reset the camera settings? May I know what SD card are you using? Did you try to reformat it?
2018-9-5
Use props
Francis
Captain
Flight distance : 774203 ft
Brazil
Offline

I have the same white dots noise issue both in 2Pro and 2 Zoom night videos. DJI must fix it.
2018-12-25
Use props
Advanced
You need to log in before you can reply Login | Register now

Credit Rules