Help me analyze these P4P waveforms... (D-LOG vs Cinelike)

Joined
Mar 20, 2016
Messages
436
Likes
161
Age
29
#1
I have four waveforms here. They all came straight from camera, went through Premiere to ProRes 422, and then had 1 node applied in Da Vinci Resolve; Color temperature. The D-LOG uses 500 ISO ("locked"), Cinelike uses 400 ISO. They are shot using same shutter speed and aperture.

As you can see, the spectrum is around as wide for all of them. I have not done any other grading or correcting other than the color temp.

The goal here is not to make any of you suprised of the result.. I came here because I can't decide for myself what is the better picture. The camera is recording my dark room with the window open, and the sun has not set yet. It is an extremly boring picture, but quite good for this test. I cannot decide which is the better picture though, they look fairly similar.

Please put the waveforms from Best to Worst, 1-4, with a comment for why you think this might be the better picture.

1: D-LOG, H265. Sharpness -3. Contrast +3
2: Cinelike, H265. Sharpness -2. Contrast 0
3. Cinelike, H264. Sharpness -2. Contrast 0
4. D-LOG, H265. Sharpness -2. Contrast +3

Skärmavbild 2017-04-25 kl. 20.31.25.png
Skärmavbild 2017-04-25 kl. 20.31.32.png
Skärmavbild 2017-04-25 kl. 20.31.40.png
Skärmavbild 2017-04-25 kl. 20.40.59.png
 
Last edited:
Joined
Nov 13, 2015
Messages
535
Likes
183
Age
63
#2
I would say 2,3,1,4
2 doesn't show any clipping
3 shows very little clipping
1 a little more clipping
and
4 a lot of clipping.

This is from my untrained and very little knowledge, but if this was video being displayed on a scope, the flat tops, to me, indicate clipping due to overdriven signal.
 
Likes: Dronefriend
Joined
Mar 20, 2016
Messages
436
Likes
161
Age
29
#3
Thanks for checking it out! I think I see what you mean. However I am leaning towards 3, 2, 4, 1 but I am not sure. The highlights in 1 and 4 have these distinct lines. Is that actual banding?

Also, the midtones and shadows in 1 and 4 might look better. But I am unsure of how to interpret them.

Look at the highlights in 3 and 2. The information is more even in 3, correct, but the info is also wider/higher. See what i mean? I think this is due to more info, and the even cut is just since the sky is not brighter than that. But again, I am not completely sure.
 
Joined
Mar 20, 2016
Messages
436
Likes
161
Age
29
#4
Alright, this is how they're done.

1: D-LOG, H265. Sharpness -3. Contrast +3
2: Cinelike, H265. Sharpness -2. Contrast 0
3. Cinelike, H264. Sharpness -2. Contrast 0
4. D-LOG, H265. Sharpness -2. Contrast +3

First of all, the -3 option in sharpness is too much. Even though the mushiness can be sharpened in POST, I don't see why I wouldn't go back to -2 and get it quite nice from the start. Who knows what details that might not be preserved by going -3. I don't know if it actually mushies what comes in, which would be bad. Otherwise the D-Logs are identical. That rules out option 1.

I find the H264 version better than the H265, even though the H265 is a superior codec. Evidently the encoder inside the Phantom is not good at handling it.

And even though I like the info better in the D-Log (number 4) for the lower half of the spectrum, we can clearly see that there is banding at the top. I was hoping that setting Contrast +3 would avoid this, since it writes to a fuller part of the available data. Apparently not. This banding is not an option.

This leaves me with another test. I have to compare D-Log h265 vs h264. If they are different and the banding is not there, D-Log might still come out as the winner, especially for low light shooting.

What a disappointment that H265 seems to be worse than H264.. but on the bright side I can work it in Da Vinci straight from the camera.
 
Joined
Mar 20, 2016
Messages
436
Likes
161
Age
29
#6
....so what's the ideal setting for cinelike now?...Is H.265 worst than h.264?
Decide for yourself bro. Test it out :) there are too many individuals on the net with varying knowledge, who tells you what is best and what not. I don't even know if I'm right.

But in this test at least, h264 produced far wider highlights than h265. Which is weird in its on way, they shouldnt differ.
 
Joined
Aug 15, 2016
Messages
1,423
Likes
295
#7
Decide for yourself bro. Test it out :) there are too many individuals on the net with varying knowledge, who tells you what is best and what not. I don't even know if I'm right.

But in this test at least, h264 produced far wider highlights than h265. Which is weird in its on way, they shouldnt differ.
What's your personal settings? DJI keeps updating/changing the color profiles, so that might be one of reasons for the diverse amount of settings here on phantom pilots.
 
Joined
Mar 20, 2016
Messages
436
Likes
161
Age
29
#8
What's your personal settings? DJI keeps updating/changing the color profiles, so that might be one of reasons for a diverse amount of settings here on phantom pilots.
This test was done on latest firmware as of today. 3840*2160 25p, save as mp4 container.

Other settings are in the OP. Focus set to manual in all tests, with same focus point.
 
Likes: tml4191
Joined
Mar 20, 2016
Messages
436
Likes
161
Age
29
#11
Alright, I have done a bigger and better test.

This test will conclude the H265 video file mode, (in 3840*2160p25 which is the only mode tested but quite possible all the other ones as well), is garbage. Not only in D-LOG but Cinelike as well. Only tested these two. Although, it cannot be proven this is because of DJIs encoder. It could be the decoder in Adobe Premiere that has some fault when it reads it. However I think it is likely to say, it is the onboard Phantom encoder.

On to the test.

This is the the test view (not my own room so no commenting on the stuff inside :D )
Skärmavbild 2017-04-26 kl. 21.08.05.png
That's a D-LOG shot by the way.

Different modes tried, using latest firmware as of this date with same shutter speed but not same aperture:

1. Normal, Contrast 0 H264
2. Cinelike, Contrast 0 H264
3. D-LOG, Contrast 0 H264
4. D-LOG, Contrast +2 H264
5. D-LOG, Contrast +2 H265

1. Normal Con 0.png
2. Cinelike H264 Con0.png
3. DLOG Con0.png
4. D-LOG Con+2.png
5. D-LOG H265.png


This time, I did not use identical ISO between non-log and D-LOG. It would not be realistic; Why would I choose to shoot in 400 ISO in Cinelike if I can choose 200 and use a more open aperture? I wouldn't.

This time, Normal and Cinelike uses ISO 200 and aperture of around 3. I tried exposing for optimal picture, like a real scenario.
D-LOG uses locked ISO of 500, and since that is brighter, I adjusted aperture somewhat, but not by much. I didn't want to lower it too much, since that would crush the blacks. I tried making the best exposure.

Then, I did a realistic correction. All pics adjusted for white balance. The Normal and Cinelike pictures got their blown highlights lowered a bit. There was a lot of info available beyond. I tried matching them to the 4. "D-LOG, Contrast +2 H264", which is untouched. Then, No 3 with Contrast 0 got stretched using Curves to match up with No 4 for better comparing.

Let's start with the very ugly result. Number 5. Look at that! Banding in the highlights, seemingly only because it was recorded in H265. No need to dwell further on this. Garbage!

1 vs 2, Normal VS Cinelike. Not sure if it is possible to see by looking at the waveforms here, but in the real picture Normal retains a bit more highlights, but Cinelike looks much more natural overall. I'd say the win goes to Cinelike. Notice how the top white line is where the highlights used to reach the limit of the scope before I pushed them down. You can see there's banding issues here as well, but this is not as surprising as in D-LOG, since this originally is the extreme of whites.

3. D-Log Contrast 0 vs 4. D-Log Contrast +2. After matching 3 with 4, I cannot tell the images from another with my eye. I think the scopes from image 4 might look better, but I am not sure. If you look closely, you can see some areas that look like start of banding in number 3, in highlights. This could be due to it being recorded to a smaller part of the available data in the codec, as opposed to 4. with +2 contrast. I will use Contrast +2 myself, for D-Log.

This leaves number 2, Cinelike and 4, D-Log. Scopewise, D-Log wins imo.

Skärmavbild 2017-04-26 kl. 21.08.05.png D-Log
Skärmavbild 2017-04-26 kl. 21.07.57.png Cinelike

But now the increased ISO becomes very apparent. The D-Log is grainier in the shadows, a lot grainier. However, it has managed to retain highlights by a lot. Everything is there! The images are recorded a minute after another.

This led me thinking: What if I recorded Cinelike in -2, as suggested by several people? Unfortunately this was around 15 minutes later, and the sky was not as bright anymore. I pushed ISO to 800: This led to blown highlights. This was perfect, since the original had that aswell. Could lowering Contrast to -2 save them from being blown, as hearsay says?
6. Conelike H264 Con-2.png
Cinelike H264, Con -2.

Turns out, it couldn't. Even though the incoming data is recorded inside the spectrum (that recording is untouched), the incoming data is blown from the start. The D-Log picture was still far superior.
This is a valuable lesson: Lowering contrast does not save any incoming data if it is already blown or crushed. It is possible, raising contrast like I did for 4. DLog +2 is irrelevant as well. But at least, in theory, it is better to save good data over a broader spectrum then a small one, if possible. But saving bad data over less spectrum won't help.

Note: I also tried number 6 in H265. Again, H265 produced more banding then H264, in Cinelike too.

Let's forget about number 6 and go back 2 & 4. Which is the winner? I don't know. What do you think?

The Conclusion:

* H265 is ****, reason unknown.
* D-Log is worse then Cinelike in dark areas because of the increased ISO, otherwise it is clearly superior.
* When you have to use ISO 400 or more, go with D-Log for better kept dynamic range.
* If you are on ISO 100, I'd go with Cinelike.
* If you are on ISO 200 it's hard to choose. But at the end of the day I hate noise. Probably Cinelike.

Personal notes: I like -2 sharpness, always. In Cinelike I'll use -1 for Saturation. in D-Log, 0.

Hope you enjoyed. Have I come to the wrong conclusion or said something incorrect? Help out and correct me.. cheers.

Edit: Damnit, the test is a bit flawed. I should have exposed D-Log higher.. with a more open Aperture.
 
Last edited:
Likes: tml4191
Joined
Jun 13, 2018
Messages
6
Likes
1
Location
Washington, D.C.
Website
www.UAVcine.com
#12
Alright, I have done a bigger and better test.

This test will conclude the H265 video file mode, (in 3840*2160p25 which is the only mode tested but quite possible all the other ones as well), is garbage. Not only in D-LOG but Cinelike as well. Only tested these two. Although, it cannot be proven this is because of DJIs encoder. It could be the decoder in Adobe Premiere that has some fault when it reads it. However I think it is likely to say, it is the onboard Phantom encoder.
After a couple of years, I found this (not much discussion about H.265, surprisingly). While the conclusion is very color-grading centric, let's keep in mind that the principal reason to use H.265 is for maintaining detail (and avoiding severe macroblocking) when there is a lot of moving content in the frame, and for that matter a lot of motion across the frame. So, when we're hovering over dense trees with lots of fluttering leaves, the improvement by using H.265 at the same bitrate as H.264 (100 Mbps for both) is absolutely dramatic, without any debate. Meantime, D-Log has gone through version upgrades, including mandatory enforcement of ISO fixed at 500, which stabilizes the conversion to REC.709 and the true dynamic range.

So in reality, unless you're shooting in low light, or forgot an ND filter and need to maintain the 180-degree shutter rule, there's no rational choice but to shoot in H.265 with D-Log in 2018.
 

New Threads

Forum statistics

Threads
129,674
Messages
1,346,784
Members
94,495
Latest member
BKSRT8