Panasonic GH5 10 bit internal recording not good enough

[Deadcode]

12 minutes ago, interceptor121 said:

That is not what I said the footage is not the point this is an analysis of the in camera codecs relative to each other 

if you can’t understand it is fine too

That is great news as I am wondering if I should buy a v90 memory card as mine fails after 30 seconds. What equipment do you use to grade? Do you have a 10 bit screen? 

when i say the codec is weak I refer to the 150 mbps. If this codec was IBP it would have been roughly equivalent to 500 mbps

 

Im grading in Davinci Resolve, and im using an old Dell U2711 (which is 8 bit), and working in REC709. 

Used to grade 5D Raw and Sony SLOG2. 

In the last couple of month i started grading in ACES, and this is where i can feel the 100Mbps 8bit  from Sony is not enough.

[interceptor121]

3 minutes ago, KnightsFan said:

You conclude with "If you want to produce genuine 10 bit colour high dynamic range footage you need to buy an external recorded capable of supporting ProRes 422 HQ or there is not game."

That's why I'm suggesting an actual comparison with Prores that uses real world examples. If you just want to compare between internal codecs, that's a misleading conclusion.

Not all Jpegs are the same, you can actually specify the quality amount and chroma subsampling of a Jpeg when you encode it. Could you specify which type of jpeg you are referring to? Also--what does jpeg have to do with any of this? I'll admit I'm having a hard time understanding why you use it as an example.

Is that a valid assumption to make, though? My understanding is that motion artifacts is one of the main reasons people use All Intra codecs.

 

To summarize my skepticism: It seems to me that you are judging codec quality purely in terms of adjusted bitrate. If I am misunderstanding, my apologies.

No you dont understand the logic of motion compression. First in order to compare static image quality you need to assume the codec is correct otherwise of course there are artefacts. The camera buffers a certain amount of frames before encoding so you are reasonably sure that the performance will be good except radical scene changes 

once you go last this point the image quality is only given by the full frames the others alone don’t even produce an image but just Delta signal to recreate one

the analysis I have done on the GOP allows me to calculate the relative size of the frames with a certain degree of accuracy although not 100% once you have  done that you can normalise your GOP to an equivalent number of full frames and desume an average for your I frames across your footage. Those values don’t change a great deal. 

Now in terms of the GH5 there are 4 codecs

all intra : no motion interpolation just saving full I framea

IPB GOP size 12 (used by all single frame rate codecs)

IPB GOP size 24 (user by 50/60p)

IP only used by 150 mbps 422 10 bit

the latter is leas efficient than the IPB so the so the size of the full frames that determines image quality drops

In terms of the conclusion that you need to use a recorder for full 422 10 bit colour this is based on the size of the I frames

of course you can’t save a jpeg with subsampling it is a static image but the impact of quality as you increase compression is similar as it is relative to itself if youn notice it or not depends on how rich is the image 

 

2 minutes ago, Deadcode said:

Im grading in Davinci Resolve, and im using an old Dell U2711 (which is 8 bit), and working in REC709. 

Used to grade 5D Raw and Sony SLOG2. 

In the last couple of month i started grading in ACES, and this is where i can feel the 100Mbps 8bit  from Sony is not enough.

Ok so the screen you use is not 10 bit how can you see the benefits?

sony 100 mbps codec is IP only and is poor compared to panasonic IPB roughly is equivalent to 170-180 mbps and is less than half the all I of the GH5

The GH5 422 10 bit 150 mbps uses IP too but has higher bitrate equivalent to 260 all I 

So as per post all-I is the best option for 10 bit and it will take some reasonable grading but shooting at 700+ will be even better assuming you can see the benefits in your setup 

 

[KnightsFan]

Thank you for the clarification. I am fairly certain now that I've been following your argument correctly. I do understand interframe compression, by the way.

However, you can't simply compare bitrates, even adjusted for motion artifacts. There are different algorithms being used. You need to show, with image analysis or an analysis of the algorithm itself, that your method of comparison is valid.

 

15 minutes ago, interceptor121 said:

In terms of the conclusion that you need to use a recorder for full 422 10 bit colour this is based on the size of the I frames

Again, for your conclusion, you have not shown that Prores is better. You are still just comparing the size of I frames from codecs that use different algorithms. That's why I suggest doing an actual analysis on real images.

15 minutes ago, interceptor121 said:

of course you can’t save a jpeg with subsampling it is a static image

Chroma subsampling is for static images, and jpegs do use chroma subsampling. "[Chroma subsampling] is used in many video encoding schemes – both analog and digital – and also in JPEG encoding." (https://en.wikipedia.org/wiki/Chroma_subsampling)

[Deadcode]

13 minutes ago, interceptor121 said:

Ok so the screen you use is not 10 bit how can you see the benefits?

With regular colorgrading it's not that hard to break an 8 bit codec. But 10 bit holds like 14 bit RAW. If your output is REC709. Probably i will step up to HDR REC2020 level in next year, maybe i will see big differences there. 

sony 100 mbps codec is IP only and is poor compared to panasonic IPB roughly is equivalent to 170-180 mbps and is less than half the all I of the GH5

The GH5 422 10 bit 150 mbps uses IP too but has higher bitrate equivalent to 260 all I 

So as per post all-I is the best option for 10 bit and it will take some reasonable grading but shooting at 700+ will be even better assuming you can see the benefits in your setup 

That's true, but for the price range and those who shoot with these consumer level cameras dont really care about the difference. Actually i want to see footage that proves the shortages compared to ProRes. 

If you are using GH5 to shoot HDR material with VLOG for production company, then yes, this matters.

 

 

[Shirozina]

You don't need a 10bit screen to see the benefits of using a 10bit codec - the lack of banding, posterization and macro blocking artefacts after applying tonal adjustments especially with LOG footage are it's big advantages. 

[interceptor121]

7 minutes ago, KnightsFan said:

Thank you for the clarification. I am fairly certain now that I've been following your argument correctly. I do understand interframe compression, by the way.

However, you can't simply compare bitrates, even adjusted for motion artifacts. There are different algorithms being used. You need to show, with image analysis or an analysis of the algorithm itself, that your method of comparison is valid.

 

Again, for your conclusion, you have not shown that Prores is better. You are still just comparing the size of I frames from codecs that use different algorithms. That's why I suggest doing an actual analysis on real images.

Chroma subsampling is for static images, and jpegs do use chroma subsampling. "[Chroma subsampling] is used in many video encoding schemes – both analog and digital – and also in JPEG encoding." (https://en.wikipedia.org/wiki/Chroma_subsampling)

There is no need to analyse the frames for a relative comparison as the AVC family of Panasonic codec use the same algorithm only the GOP structure changes this is well explained in panasonic documentation of the codec so the assumption holds true

in terma of avc intra being a proxy for prores I take the point but I have no information to conclude that the image processing of avc intra is better

 

[KnightsFan]

21 minutes ago, interceptor121 said:

There is no need to analyse the frames for a relative comparison as the AVC family of Panasonic codec use the same algorithm only the GOP structure changes this is well explained in panasonic documentation of the codec so the assumption holds true 

I think you misunderstand me. We all know from experience that the quality of interframe compression varies based on the amount of motion in the frame. You even showed that as an example with the water fountain. I think it is faulty logic to assume that you can come up with a generic file size for an I frame, out of a long GOP file. I'll try to explain myself more clearly:

 

"With an average video bit-rate of 94 Mbps each GOP has 45.3 Mbps"

So far so good. We have 94 Mbits to describe a second of footage.

"which means an I Frame has around 13.1 Mbits or 1.57 MB per frame"

Ok, that's true. I can see from your chart that the I frame is 1648326 bytes, so that I-frame is indeed ~1.6 MB.

"and an equivalent All-Intra bit-rate of approximately 328 Mbps."

This is where I disagree. You are apparently multiplying the 13.1 Mb/frame by 25 frames/sec to get the "equivalent" data rate of 327.5 Mbps. You are assuming that the amount of data in a B frame (247334 bytes for example) is the SAME amount of data needed to retain the fidelity of that frame that is required by the I frame in total. That is, you assume that:

1648326 absolute bytes = 247334 delta bytes

Where "equivalence" means that they will retain the same amount of information relative to the original, real world scene that is being encoded.

And we know, based on your example of the water fountain, that interframe compression quality depends heavily on what is being shot. Therefore, the idea of "equivalent" data rates being used as a measure of actual information retention is faulty, unless you specify what type of data is encoded.

 

34 minutes ago, interceptor121 said:

in terma of avc intra being a proxy for prores I take the point but I have no information to conclude that the image processing of avc intra is better

I actually think you are right about Prores retaining more information, but like I said you don't have evidence for it. And beyond that, the GH5 could have such poor image processing that uncompressed HDMI actually has no benefit over the internal recording either so even if Prores is theoretically better, concluding that people need an external recorder to get 10 bit quality does not seem justified.

[interceptor121]

1 hour ago, KnightsFan said:

I think you misunderstand me. We all know from experience that the quality of interframe compression varies based on the amount of motion in the frame. You even showed that as an example with the water fountain. I think it is faulty logic to assume that you can come up with a generic file size for an I frame, out of a long GOP file. I'll try to explain myself more clearly:

 

"With an average video bit-rate of 94 Mbps each GOP has 45.3 Mbps"

So far so good. We have 94 Mbits to describe a second of footage.

"which means an I Frame has around 13.1 Mbits or 1.57 MB per frame"

Ok, that's true. I can see from your chart that the I frame is 1648326 bytes, so that I-frame is indeed ~1.6 MB.

"and an equivalent All-Intra bit-rate of approximately 328 Mbps."

This is where I disagree. You are apparently multiplying the 13.1 Mb/frame by 25 frames/sec to get the "equivalent" data rate of 327.5 Mbps. You are assuming that the amount of data in a B frame (247334 bytes for example) is the SAME amount of data needed to retain the fidelity of that frame that is required by the I frame in total. That is, you assume that:

1648326 absolute bytes = 247334 delta bytes

Where "equivalence" means that they will retain the same amount of information relative to the original, real world scene that is being encoded.

And we know, based on your example of the water fountain, that interframe compression quality depends heavily on what is being shot. Therefore, the idea of "equivalent" data rates being used as a measure of actual information retention is faulty, unless you specify what type of data is encoded.

 

I actually think you are right about Prores retaining more information, but like I said you don't have evidence for it. And beyond that, the GH5 could have such poor image processing that uncompressed HDMI actually has no benefit over the internal recording either so even if Prores is theoretically better, concluding that people need an external recorder to get 10 bit quality does not seem justified.

I have looked into Panasonic avc intra documentation

In terms of all intra as implemented in the GH5 Cabac entropy encoding is not used only DCT is used so the assumption that this implementation of avc intra is similar to prores holds 

For what concerns the equivalent bit rate calculation that assumes no error of motion interpolation. This of course can be true or not depending on the situation however on a static scene the calculations hold and the considerations are valid furthermore IPB codec use cabac encoding so provided that the motion interpolation works it is likely that the quality of the image is pretty much the same in the 420 8 bit implementation (Panasonic claims 20-30% increased efficiency with cabac)

It is obvious that the choice between an intra codec and long gop depends on the amount of motion in the scene and what this proves is that the 100 mbps and 400 mbps produce very similar results in certain circumstances. 

The post only aims at ranking the various options and to show that one of the implementations the 150 mbps 422 10 bit long gop is inferior to the others as the h264 codec is the same and the settings for that one are worst than the others

prores conclusions need testing and it is possible that the camera is not good enough to push the boundaries of Prores 422 HQ anyway and therefore we see no difference from all intra

 

2 hours ago, Shirozina said:

You don't need a 10bit screen to see the benefits of using a 10bit codec - the lack of banding, posterization and macro blocking artefacts after applying tonal adjustments especially with LOG footage are it's big advantages. 

Am not sure I follow fully some of those issue depend on the quality of the codec not just 8 to 10 bits

8 bit colour should not give banding anyway and the fact you can apply corrections without breaking the codec has to do with the compression headroom more than anything else?

[newfoundmass]

I dunno, I am pretty happy with the 10 bit I get internally... You're welcome for my deep analysis! 

[OliKMIA]

Yes +1. The 10 bits is nice to have but the bitrate is too limited for 4k video whether it's 150 mbps Long GOP or 400 All-I.
I never really understood the point of 4k/10bits footage recorded in h.264 with limited bitrate. Personally I prefer to use 4k/60 in 8bits instead of 4k/30 in 10bits. But at least Panasonic proposes the option and some people really like it. The GH5 is a great camera.

[Benjamin Hilton]

I have shot extensively with the GH5 cameras over this last year and I can confirm that there is a night and day difference between the internal 8bit 100mb/sec and 10bit 150mb/sec in V-Log.  Even using Cine-Like D is a big difference.  If you are shooting natural or rec 709 though, the 8bit 100mb/sec holds up great.  It just depends on what you are shooting.  S-log 2 is made to hold up with Sony's 8bit codecs, V-log is not. 

[jonpais]

@Benjamin Hilton But it’s no good at all. Don’t judge with your own eyes - look at the numbers I’ve crunched on this spreadsheet! ?

[Shirozina]

6 hours ago, interceptor121 said:

Am not sure I follow fully some of those issue depend on the quality of the codec not just 8 to 10 bits

8 bit colour should not give banding anyway and the fact you can apply corrections without breaking the codec has to do with the compression headroom more than anything else?

Yes that's true (The HDMI out on my Sony A7s and A7r2  is very robust ) but it still doesn't change the fact that you don't need a 10 bit screen to benefit from a 10bit codec. I have argued many times previously that 10bit is not a cure-all. The advantage in video of 10bit is that in a YCBCR codec most of the data is stored in the Luma channel so higher rates of compression can be applied without running into some of the more obvious tonal artefacts. This I assume is why the 'theoreticaly inferior' 150mbps 10 bit codec can comfortably be used to shoot V-LOG, be subject to lots of tonal correction and still come out looking very good where as most internal 8 bit codecs fall apart.

At the end of the day most users are not choosing their  codecs after doing theoretical data analysis exercises as it's not hard to see the artefacts of a poor codec once you start grading esp if you shoot LOG profiles.

 

[interceptor121]

1 hour ago, Shirozina said:

Yes that's true (The HDMI out on my Sony A7s and A7r2  is very robust ) but it still doesn't change the fact that you don't need a 10 bit screen to benefit from a 10bit codec. I have argued many times previously that 10bit is not a cure-all. The advantage in video of 10bit is that in a YCBCR codec most of the data is stored in the Luma channel so higher rates of compression can be applied without running into some of the more obvious tonal artefacts. This I assume is why the 'theoreticaly inferior' 150mbps 10 bit codec can comfortably be used to shoot V-LOG, be subject to lots of tonal correction and still come out looking very good where as most internal 8 bit codecs fall apart.

At the end of the day most users are not choosing their  codecs after doing theoretical data analysis exercises as it's not hard to see the artefacts of a poor codec once you start grading esp if you shoot LOG profiles.

 

LOG is another form of compression overlayed on all the rest. I don’t use it but I would think this is too much to take for the 100/150 IPB codecs maybe the 150 mbps IP only works better with it assuming you don’t get motion artefacts

For what concerns Luma my understanding is that it is linked to RGB coefficients it does not exist in isolation so your greys come fromntbeyae source so if you cant display 10 bit red you won’t see it either for me banding etc are more due to compression. I had a sony camera and the codec was the same at avchd and xavc despite different container I would get banding with 24 mbps but not with 50 and it was still 8 bits

LOG should work better with All-Intra as there is more headroom

just for clarity am not saying this camera is crap what am saying is that some of the features are not fully implemented at least for in camera recording

Maybe I should have put a question mark in the title...

[deezid]

10 hours ago, OliKMIA said:

Yes +1. The 10 bits is nice to have but the bitrate is too limited for 4k video whether it's 150 mbps Long GOP or 400 All-I.
I never really understood the point of 4k/10bits footage recorded in h.264 with limited bitrate. Personally I prefer to use 4k/60 in 8bits instead of 4k/30 in 10bits. But at least Panasonic proposes the option and some people really like it. The GH5 is a great camera.

Just shoot a clear sky in V-log using the 8 bit codecs.

 

And you'll never do it again. Try using the codec instead and do colorgrading in Resove.

 

In terms of actual visible quality the GH5 codecs behave like this:
400Mbit/s Intra and 150Mbit/s IPB 10bit both look visually the same in most situations and grade extremely well, even in extreme low light with lot's of noise and usually don't show any banding artifacts. I actually managed to create banding using these codecs only once since I've bought the camera (gray wall underexposed).

200 Mbit/s 10bit 4:2:0 H265 codec for 6K anamorphic has some slight banding issues and shows some macro blocking.

150Mbit/s 8bit codec doesn't like gradients at all. Workaround is to use CineLike D and the Cinelike D to V-log conversion by @Sage

72Mbit/s 10bit H265 codec shows some slight banding, macro blocking and even some blur.

100Mbit/s 8bit codec doesn't work with gradients nor noise. Comparable to the GH4 H264 implemenation. Don't ever use it.

 

Some showcase of the terrible 150mbit/s 10 bit codec:

 

[no_connection]

Just record externally uncompressed and then compare it with internal and show the actual difference in the image.

Comparing bitrate to prores isn't even beginning to be useful, especially since prores is, how do I say this nicely, kinda crap when it comes to bitrate efficiency. prores shoves pitrate at the problem to compensate for being easy on the CPU to decode.

in H264 there is a huge difference in the result depending on how much effort you put into encoding it. For example using same bitrate a fixed function encoding chip could produce kinda crappy image where a CPU using maximum effort could produce a nice looking image.

[deezid]

3 minutes ago, no_connection said:

Just record externally uncompressed and then compare it with internal and show the actual difference in the image.

Comparing bitrate to prores isn't even beginning to be useful, especially since prores is, how do I say this nicely, kinda crap when it comes to bitrate efficiency. prores shoves pitrate at the problem to compensate for being easy on the CPU to decode.

in H264 there is a huge difference in the result depending on how much effort you put into encoding it. For example using same bitrate a fixed function encoding chip could produce kinda crappy image where a CPU using maximum effort could produce a nice looking image.

yep, that's exactly how it is.
The 10 bit 150mbit/s implementation is way more effective but also more CPU excessive than the other H264 implementations. And in terms of artifacting it surpasses every 8 bit H264 choice on this camera quite easily in basically every situation (Talking about V-log L here).

[MurtlandPhoto]

Not a technical comparison, but these guys saved me the time to do my own testing months ago. Haven't looked back since ?

 

[OliKMIA]

2 hours ago, deezid said:

Just shoot a clear sky in V-log using the 8 bit codecs.

And you'll never do it again. Try using the codec instead and do colorgrading in Resove.

Sure, it's not completely useless and but the bitrate and codec are very limiting. But again, it's nice to have.

[Dan Sherman]

It's better than 99% of the "technical" and "mathematical" reviews I've seen by people who are using pseudoscience. It seems like about 1 in 1000 understand their is more than bit rate, bit depth, chrome sampling, and codec involved in getting a good image.  

It hilarious how many individuals think they more more than a large huge company.
 

1 hour ago, MurtlandPhoto said:

Not a technical comparison, but these guys saved me the time to do my own testing months ago. Haven't looked back since ?