h265 potential banding issue?

[Tomas Wangen]

Note that this is by no means conclusive - one scene sample only.

When comparing h264 / h265 for a complex scene (downward movement with lots of trees and full sun) I noticed that the h265 picure shows clear indications of banding in the ocean which did not show up for the same h264 footage (all settings the same otherwise 4k / 30 / none profile all manual).

Cranking up the shadows, this is what I saw every second; h264 had almost constant quality while h265 pulsed every 30 frames with high quality gradually degrading (clear pulse visible in the scopes).

Again, this is the only test scene where this was obvious, it would however be interesting if others have seen the same or could repeat this.

Rain this weekend, so no more footage from me..

Note that is has a extreme curve to show the issue

Edit: This video is a crop of the source material, both extreme curve and normal. You should be able to see much more pronounced banding in the h265 parts, and the pulse should also be evident.

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

 

[Zzz]

Interesting... is there a technical explanation for the pulsing / banding?

 

[MarcelV]

From what I have read and seen elsewhere this is due to a problem with the h265 compression as implemented by DJI. All of these MPEG-4 based codecs use a GOP (group of frames) compression scheme where the full data is stored for the first frame in the group (called an I frame) and then from frame to frame only the changes from the previous frame are stored. With large GOPs (more than 15 frames or so) it becomes increasing likely that image degradation will show up as more and more of the image is being interpolated late into the groups frames (small losses from frame to frame get compounded).

For the DJI camera it seems they are setup for either 30 frame GOPs or perhaps GOP = to fps (not sure as I am new to the drone cameras). In the case of the P4P and some of the inspire cameras the implementation causes a loss of brightness and detail as it progresses to write the group, which is re-freshed every second when it starts a new group and records a new full quality I frame.

That is something of a simplification, but this will need to be fixed in firmware. Hopefully soon or it really limits the usefulness of the h265 feature.

 

[smiller]

I'm wondering, if all these issues are endemic to DJI products (and it seems like they would be since DJI would use the same base code for these types of functions) then how are Inspire cameras, Osmos, etc. suitable for pro or semi-pro use?

 

[MarcelV]

I'm wondering, if all these issues are endemic to DJI products (and it seems like they would be since DJI would use the same base code for these types of functions) then how are Inspire cameras, Osmos, etc. suitable for pro or semi-pro use?

Well, the "pros" don't want you to know this, but lots of lower end codecs are used to acquire footage that ends up on TV etc. In the case of drones the reduced cost of not hiring a full on helicopter and Space-Cam setup far out ways any "lower quality" footage concerns.

The new Inspire 2 can record directly to ProRes 4444 (lossless) on an SSD which (should) bring the quality level of footage to a new level (and the cost to acquire as well).

 

[smiller]

That may be true, but it begs the question... what should we expect out of the P4P camera system? It is part of an expensive RC aircraft and controller combo so out of the total price how much is left for the camera? It is demonstrably better than the P4 and predecessors but yes, falls short of an Inspire 2 or ARRI Alexa. There is the reality of a price point after all.

 

[Tomas Wangen]

It seems that for now, we should probably stick to none-colorprofile with no contrast adjustment and h264 as well as make sure we have a perfect exposure (exposed to the right) and we will have very good results in 99% of the cases.

I hope they improve the h265 settings (as h265 should be much better compared to h264 given the same bandwidth) and also give us a working log profile to improve our options in difficult conditions.

 

[KevMo Photog]

Def following all this thread as I am also concerned with video quality big time. So is this issue with the h265 codec something that could possibly be fixed in a firmware update?

 

[Tomas Wangen]

Banding might be a issue with low-bandwidth h265 in general, the 30 frames between i frames seems unneccesary. There are a lot of encoder options, so let us see if dji can come up with a improvement. I assume someone from dji is following this forum?

I will stick with h264 for now.

 

[smiller]

I assume someone from dji is following this forum?

I wouldn't count on that, the DJI forum is a much better place to report issues. The users there don't seem to be nearly as technically competent as here and the thread quality sometimes isn't very good, but DJI reps will directly interact with problem reports so it is a much better way to get noticed.

You can also open a ticket by sending a report to [email protected], attaching any documentation or examples you might have. If the files are large use a Dropbox link, that's what they seem to prefer for larger submissions.

 

[Dronefriend]

Tomas. Great testing on your part as with many others on this site. Do you know if the issue is persistent in 24 and 25 fps as well?

 

[Tomas Wangen]

I never shoot below 30 fps, so this is unknown. However, 24 fps should give more headroom given the same bitrate so it might have fewer encoder-related issues.

If anyone would like to test, downward/upward motion with lots of trees combined with a uniform area, preferrably with contrasty light) should yield easily comparable (and repeatable) results.

 

[MarcelV]

That may be true, but it begs the question... what should we expect out of the P4P camera system? It is part of an expensive RC aircraft and controller combo so out of the total price how much is left for the camera? It is demonstrably better than the P4 and predecessors but yes, falls short of an Inspire 2 or ARRI Alexa. There is the reality of a price point after all.

Sure, I agree that we are getting a nice camera perhaps cheaper than we would if it were not bundled with a aerial platform. But if the software encoding is not up to par with the quality of images the camera is capable of, then I do have an issue with that.

It should be correctable via firmware, but I don't know if they will actually correct it.