DV Info Net

DV Info Net (https://www.dvinfo.net/forum/)
-   Panasonic LUMIX S / G / GF / GH / GX Series (https://www.dvinfo.net/forum/panasonic-lumix-s-g-gf-gh-gx-series/)
-   -   Gh5 firmware 2,0 (https://www.dvinfo.net/forum/panasonic-lumix-s-g-gf-gh-gx-series/534721-gh5-firmware-2-0-a.html)

Cary Knoop October 5th, 2017 08:38 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Cliff Totten (Post 1937125)
Now, I tested my camera on the original GH5 firmware a while back.

1.0 definitely had some issues (ghosting) with v-log, issues that were fixed in 1.1

I don't know if any things were fixed (under the covers) for 2.0, but I do like the colors coming from the 4992 x 3744 10 bit 4:2:0 V-Log encoded with H.265.

Gary Huff October 5th, 2017 09:09 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Cliff Totten (Post 1937125)
part 3 - is a color canceled luma grayscvale filter. This is "looking under the hood" of the CODEC. You can see the Long GOP IPB cadence at work. Oddly enough, there are some straight line and block artifacts on the left side of the image but it's fully covered up under normal RGB viewing.

This is not actually a thing. You can use a channel splitter which takes YUV and splits it into what the RGB components are, but you will not see the LongGOP IPB "cadence" (again, not a thing). In fact, I'd say what you're calling the "banding" (it's not banding), is actually some kind of render error from whatever filter your adding to that. Something in AVISynth?

EDIT: Yes, when I split out the signal into individual RGB components, the lines from that rendered aren't there for me in Resolve Studio 14. I'm going to say that since you're trying to concoct a LUT via LUTCalc, that your issues are stemming from that.

Gary Huff October 5th, 2017 09:28 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Cary Knoop (Post 1937126)
I don't know if any things were fixed (under the covers) for 2.0, but I do like the colors coming from the 4992 x 3744 10 bit 4:2:0 V-Log encoded with H.265.

The ghosting was at high ISOs, which was probably some kind of software error involving high ISO NR.

None of the release notes mentions touching any modes that already existed, simply unlocking new ones, so I highly doubt there are any changes to the original recording modes.

Cary Knoop October 5th, 2017 09:51 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Gary Huff (Post 1937128)
None of the release notes mentions touching any modes that already existed, simply unlocking new ones, so I highly doubt there are any changes to the original recording modes.

4992 x 3744 10 bit 4:2:0 V-Log encoded with H.265 is new.

Cliff Totten October 5th, 2017 09:54 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Gary Huff (Post 1937127)
This is not actually a thing. You can use a channel splitter which takes YUV and splits it into what the RGB components are, but you will not see the LongGOP IPB "cadence" (again, not a thing). In fact, I'd say what you're calling the "banding" (it's not banding), is actually some kind of render error from whatever filter your adding to that. Something in AVISynth?

EDIT: Yes, when I split out the signal into individual RGB components, the lines from that rendered aren't there for me in Resolve Studio 14. I'm going to say that since you're trying to concoct a LUT via LUTCalc, that your issues are stemming from that.

So that gray scale effect happens from a mixture of the 6 primary and secondary colors. (red, yellow, green, cyan, blue and magenta) The unsaturated parts are also included. This phase cancels out the color and brakes the CODEC in a neat way.

When I say you can see the IPB, what I mean is that you can actually see MPEG working long GOP calculations and plotting all around the image. This kind of phase cancellation reveals elements you cant normally see easily. This stuff is always somewhat hidden under normal circumstances. It's kind of useless but when you do this to ProRes HQ, you dont get artifacts like this. This is where complete I frames are superior. They dont need any data from previous frames to assemble a solid image. It's interesting to note that ProRes is a DCT-type system and is actually a relative of JPEG!

LutCalc wouldn't do those line artifacts to the image.

CT

Gary Huff October 5th, 2017 09:59 PM

Re: Gh5 firmware 2,0
 
What is your exact workflow for turning an image to grayscale? How does this process, by simply turning the image grayscale, invoke more information than splitting the video into individual R, G, and B elements and viewing each independently?

Doing a splitter node in Resolve does not yield any difference between LongGOP and Intra as you watch each channel isolated.

Gary Huff October 5th, 2017 10:01 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Cliff Totten (Post 1937130)
When I say you can see the IPB, what I mean is that you can actually see MPEG working long GOP calculations and plotting all around the image.

Please give a single, specific example of where one can actually see these GOP calculations happening in the video you uploaded.

Gary Huff October 5th, 2017 10:08 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Cary Knoop (Post 1937129)
4992 x 3744 10 bit 4:2:0 V-Log encoded with H.265 is new.

Yes, its a new mode that has been added, not a previously existing mode that has been updated.

Not really sure Id ever want to shoot in that, being 4:2:0.

Cliff Totten October 5th, 2017 10:27 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Gary Huff (Post 1937132)
Please give a single, specific example of where one can actually see these GOP calculations happening in the video you uploaded.

So you see pixel clusters and shapes moving around the flat, non detailed portions the black and white image I uploaded? This is part of the Long GOP process. It's recycling "crumbs" from frames before each other.

And no, before you complain..."crumbs" is not a legal, technical, or "official" MPEG Long GOP term. It's my very own created word to describe the data carry over.

Haven't you ever taken a ProRes HQ master file, then compressed it into H.264 or MPEG 2 and re-applied that new copy (frame sync'd) on top of the original master with a phase inverted filter to reveal (in a gray scale) the Long GOP artifacts that the original master does not have? I'm pretty sure everybody has done this before for fun.

CT

Gary Huff October 5th, 2017 10:30 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Cliff Totten (Post 1937135)
Haven't you ever taken a ProRes HQ master file, then compressed it into H.264 or MPEG 2 and re-applied that new copy (frame sync'd) on top of the original master with a phase inverted filter to reveal (in a gray scale) the Long GOP artifacts that the original master does not have?

What is your workflow exactly for applying this phase inverter filter? What is the software you are using and what is the filter?

Cliff Totten October 5th, 2017 10:40 PM

Re: Gh5 firmware 2,0
 
Any NLE can invert a waveform. You are flipping the wave "inside out" Vegas 14, Premiere, anything.

Take two virtually identical sine waves. Make even the slightest frequency change or phase shift to one, then invert it and apply back it to the other. The parts that are in phase will disappear because of cancellation. They only parts that exist will be the EXACT elements that were altered in the second one.

It's called "phase cancellation". Even phase canceling color channels and blending the inverted waveform will wipe out wave shapes and reveal others.

I know that a man as smart as you Gary, already knows this. So why are you asking me to explain it to you?

Time for bed!

CT Zzzzzzzz

Cary Knoop October 5th, 2017 10:41 PM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Gary Huff (Post 1937133)
Yes, it’s a new mode that has been added, not a previously existing mode that has been updated.

Not really sure I’d ever want to shoot in that, being 4:2:0.

What is the problem with that?

This "open gate" 5K mode takes almost the full sensor resolution. The full sensor has just as much (but differently formatted) information as 4:2:0, there is simply not more information available.

When you scale 5K 10 bit 4:2:0 video to 75% to make it UHD you get an approximate equivalent amount of information to 4:2:2. Theoretically you could even improve the bit resolution!

Plus you have more flexibility in framing in case you need to because you record in 4:3! And 2.0 offers aspect ratio markers in display!
Also this mode effectively gives you an equivalent 400Mbps Long GOP mode!

This is definitely my favorite mode to record in, and the colors look fantastic!

If you have not tried it I highly recommend it!

Gary Huff October 5th, 2017 10:47 PM

Re: Gh5 firmware 2,0
 
Actually, you can’t. Better to let the GH5 do the processing. There is no contest between recording 4K 4:2:0 and downscaling into 1080 (for the theoretical 4:4:4) and capturing in 444 RGB mode. The real world results simply aren’t there.

Also, bit precision absolutely will not increase with a downscale at all. That’s done at the capture level, even before debayering.

Cary Knoop October 6th, 2017 12:29 AM

Re: Gh5 firmware 2,0
 
Quote:

Originally Posted by Gary Huff (Post 1937139)
There is no contest between recording 4K 4:2:0 and downscaling into 1080 (for the theoretical 4:4:4) and capturing in 444 RGB mode.

There is nothing theoretical about it.

4:2:0 4K color (CbCr) information is already at 2K, so by just down scaling the luminance (Y') 50% you automatically get 4:4:4. What is incorrect is that the bit depth increases 4 times as well (due to oversampling), actually only the luminance component increases the bit depth with a factor 4, the color info actually remains at 8 bit because it is actually untouched with the down scaling.

Feel free not to use the 5k mode, I think you miss a lot of opportunities. :)

Cliff Totten October 6th, 2017 06:11 AM

Re: Gh5 firmware 2,0
 
Yes, the open gate mode is awesome and even at 4:2:0, you still have a huge amount of chroma resolution. However, the bit depth cant change. I see what you are saying about pixel scaling but whenever you take two or more 10bit values and mix them or sum or bin them, you will always get a new 10bit value.

Now if you took one pixel with a 10bit value and combined it with another pixel that had a 12bit value then you could achieve a new 12bit sum. But you could mix and match all the 10bit values you want together but you will always end up with a new 10bit product.

Here is another way to visualize this. Imagine millions of pixels but each one having only a 3 color scale. (3 color shade possibilities)

0 = pure black
1 = middle grey
2 = pure white

You could mix millions of these 0, 1 and 2 pixels as much as you want in a 32bit color space but you will never get any shade more than black, grey and white. You would need to introduce a value that is "outside" of that 3 shade scale to offset these 3 numbers to produce values that are beyond 0,1 and 2. (like tossing in a "1.3" grey pixel or a "0.5" grey or a "1.9" white pixel...but those pixels you just dont have)

But yeah, I have seen some fantastically detailed videos shot with the full 4:3 sensor. Check this one out its CRAZY detail. You can literally make out faces of people way off in the distance.



All times are GMT -6. The time now is 08:52 PM.

DV Info Net -- Real Names, Real People, Real Info!
1998-2024 The Digital Video Information Network