Color mode & Dynamic Range

[yannick_c]

Thinking of getting my P4P back to the store. Camera is great, lot of details, ... but that horizon drift when moving sideway is killing me (and the shots). Maybe in 1 or 2 months, everything will get ok and we'll have great profiles and great gimbal behavior (even I doubt on that last thing)

 

[F3honda4me]

Thinking of getting my P4P back to the store. Camera is great, lot of details, ... but that horizon drift when moving sideway is killing me (and the shots). Maybe in 1 or 2 months, everything will get ok and we'll have great profiles and great gimbal behavior (even I doubt on that last thing)

My horizon stays perfect when moving sideways.

 

[David_Cambridge]

Andrei V

The cinelook is functioning exactly as it should. There is massive confusion about what increased dynamic range means. By design, it expands the range of light and darks in the subject that can be recorded to the 256 bit range that is mostly used for video (8 bits). For a subject of average luminance range it compresses it into fewer bits so that there is more space to encode the increased luminance range of the SUBJECT. This means that the average subject will be compressed in the video waveform. That is what cine setting does. It is almost identical to reducing the contrast using the contrast control.

It can easily be expanded back to a full range video with any common post processing tool.

Dlog is badly broken because it does compress the luminance range of the original subject as designed, but includes the original clipping behavior as well. This is wrong. By expanding the capture range. the resulting image should fit within the 8 bit video without being clipped. That's the whole point of expanding the range.

 

[Raptorman0909]

If the weather isn't so bad tomorrow I plan to do some flying tomorrow with my replacement P4P. I don't know that I'll test all the things that have been changed but will likely give D-Log another look even though it appears that not much has changed with DJI's bad implementation of it. I had decent results with D-Cinelike and will give that a go for sure, but I don't know that I'll bother with any of the new Film-X profiles. I also plan on using a custom setting of -2/0/-2 and probably also 0/0/0.

To get more than that I'd want to test from the ground with a high contrast subject and vary the exposure from under-exposed to over-exposed using some of the profiles and custom settings.


Brian

 

[Andrei V]

Andrei V

The cinelook is functioning exactly as it should. There is massive confusion about what increased dynamic range means. By design, it expands the range of light and darks in the subject that can be recorded to the 256 bit range that is mostly used for video (8 bits). For a subject of average luminance range it compresses it into fewer bits so that there is more space to encode the increased luminance range of the SUBJECT. This means that the average subject will be compressed in the video waveform. That is what cine setting does. It is almost identical to reducing the contrast using the contrast control.

It can easily be expanded back to a full range video with any common post processing tool.

Dlog is badly broken because it does compress the luminance range of the original subject as designed, but includes the original clipping behavior as well. This is wrong. By expanding the capture range. the resulting image should fit within the 8 bit video without being clipped. That's the whole point of expanding the range.

With all due respect, David -- have you actually tested a P4P in dcinelike as compared to the none profile? Or experimented with negative contrast in the custom mode?

While what you describe is theoretically sound and what should be happening if dcinelike was actually implemented in the same way as similar modes on other cameras, it's not what's actually happening in practice using the current DJI implementation of dicinelike on the P4P.

I have tested dcinelike in scenes where there is highlight clipping or shadow clipping using the None color mode. Switching to dcinelike with the exact same exposure doesn't capture any additional detail in the clipped areas as far as I can tell. Expanding the waveform in post just gets you back to where you would've been in the first place filming in None.

In other words, Dcinelike appears to be broken in the exact same fashion Dlog is broken. Dcinelike includes the original clipping behavior.

Here are the shots:

 

[Moonfox]

If DJI has an awesome Sony sensor in the p4p, wouldn't Sony have awesome profiles to go with it? Or would the p4p lens and camera charecteristics make them useless? Or if not, Sony wouldn't be inclined to share (or license) to DJI? Or DJI not willing to pay?

 

[David_Cambridge]

You are correct. I have not tested it in a p4. I have tested in in my p3 4K where it works as advertized. I use it all the time.

Sorry about the mis-information.


Sent from my iPad using PhantomPilots

 

[David_Cambridge]

Good plan. Note that my earlier remarks may be applicable only to the P3 4K. It appears that DJI changes stuff even using the same camera. What a mess.

 

[Dronefriend]

I feel like we are not talking the same language here. D cinelike does NOT expand 0-15 IRE to say 0-25 IRE by compressing 25-40 ish IRE midtones. It simply pushes 0-10 up, but the section of darks is still the same width. But now the midtones are suffering from less width. This is what we have been saying all the time. If the P34k behaves differently I want proof like the screenshots that has been provides here, because I dont believe it.

Edit: this upsets me because you keep on saying how using a negative contrast will yield other good results, but you are completely wrong. Just as in the cinelike scenario, lowering contrast does not yield a broader spectrum by allowing broader lows and highs for narrower mids: it simply compresses everything. This is clearly demonstrated in every histogram posted from such tests. Outside of the compressed area there is no information at all. If you were right the spectrum would still be full from 0-255 even though the midtones were compressed, but I repeat, the extremes are left completely blank. It does not protect you from clipping, it just put the clipping farther to the midsection.

 

[yannick_c]

I don't agree with you all
For me, as I said earlier on my first tests, D-Cinelike keeps more infos/details in the shadows than None (but a little less on the highlights). None clips before D-Cinelike in the shadows.

An example : Video taken in Manual, at the same exposure, same parameters (0,0,0), same everything ... only changed from None to D-Cinelike. Then in Lumetri, I added +2ev. Just look how D-Cinelike keeps more details in the shadows (not that much, but it's not the same) where None shows lot of artifacts

 

[yannick_c]

Same thing but with +4ev. OK it's horrible, but there's more details

 

[smiller]

OK so after reading the thread I'm now convinced that D-Cinelike is either useless, or helpful

 

[yannick_c]

The thing is that it's kind of 0.15 stop more in the shadows, so when compared to the ~11.5 stops of the P4P, it's just 1% more, so not really that good. We won't really see the difference at the end.

 

[Tomas Wangen]

OK so after reading the thread I'm now convinced that D-Cinelike is either useless, or helpful

Lifting shadows will always introduce artifacts regardless of mode; dCineLike might be slightly more forgiving but in no way able to restore information below 0. It is normally better to sacrifice highlights (as around the sun) than shadows and the rule in general is to stay in the top spectrum as much possible (ideally without clipping).

Exposing to the Right Explained

 

[2nd2non]

For aerials, it's generally more important for me to preserve shadows w/o adding artifacts. Sounds like d-cinelike is a little more beneficial for this purpose


Sent from my iPhone using Tapatalk

 

[Dronefriend]

I don't agree with you all
For me, as I said earlier on my first tests, D-Cinelike keeps more infos/details in the shadows than None (but a little less on the highlights). None clips before D-Cinelike in the shadows.

An example : Video taken in Manual, at the same exposure, same parameters (0,0,0), same everything ... only changed from None to D-Cinelike. Then in Lumetri, I added +2ev. Just look how D-Cinelike keeps more details in the shadows (not that much, but it's not the same) where None shows lot of artifacts

What Phantom, bitrate and resolution is this footage from?

If this is from your shoot with the P4P 4K @ 100mbit then I agree with you that DCinelike is preserving more shadows. Although not as much as it should. I think you may be right about it being the way to go though. I will have to test it myself tomorrow evening. Good that you stayed to your point. This doesn't change the fact that using a -contrast mode will yield nothing, though. And, as I've stated, DJI should have implemented the feature better than this.

 

[Tomas Wangen]

For aerials, it's generally more important for me to preserve shadows w/o adding artifacts. Sounds like d-cinelike is a little more beneficial for this purpose


Sent from my iPhone using Tapatalk

Actually to preserve the shadows you only have to watch your histogram and make sure you do not clip the shadows; again mode itself is not affecting this much, if any.

 

[Dronefriend]

If it is correct that DCinelike can preserve more shadows, I suspect it's due to the codec being bad at preserving information at the lowest blacks 0-5 IRE, and that it benefits from having these pushed up to over 5ish IRE. It is not that the mode uses a different curve for information storing. This means the mode does what it's supposed to but for the wrong reasons. Either way, if this is true I will still personally use it to avoid having info at the lowest black where the codec may not preserve it well.

 

[yannick_c]

What Phantom, bitrate and resolution is this footage from?

If this is from your shoot with the P4P 4K @ 100mbit then I agree with you that DCinelike is preserving more shadows. Although not as much as it should. I think you may be right about it being the way to go though. I will have to test it myself tomorrow evening. Good that you stayed to your point. This doesn't change the fact that using a -contrast mode will yield nothing, though. And, as I've stated, DJI should have implemented the feature better than this.

Yes, P4P 4K h265 100Mbps. That was not on the last firmware but not sure they change D-Cinlelike between the previous firmware and the last one.

 

[Tomas Wangen]

Yes, P4P 4K h265 100Mbps. That was not on the last firmware but not sure they change D-Cinlelike between the previous firmware and the last one.

I actually had some interesting findings regarding h265 (as compared to h264) and shadow areas:

h265 potential banding issue?