GH5 Game Over, Panasonic

[combatentropy]

For all practical purposes when refering to LOG vs RAW (and especially when comparing a camera that can record RAW vs a camera that can't) it is understood that we are talking about using LOG compression to represent a wider dynamic range within the narrower signal of an RGB or YUV signal, not, of course, the broad mathematical concept of a logarithmic curve. That would be uselessly academic.

Log isn't compression.

As John Brawley pointed out, there are cameras whose RAW format is logarithmic. Logarithmic has been a mode in scientific and industrial cameras for a long time. Log recording dates back at least to Kodak's Cineon system, from the early 90s. I first heard about it with the Thomson Viper Filmstream camera (NAB 2002) famous for its uncompressed RGB log output. Whether you're recording compressed or uncompressed, 8-bit or 12-bit, I believe log is a better use of of your bits.

Linear is wasteful. Look at how many levels are given to the upper f-stops and how few to the lower ones, even with 12 bits:

F-stop	Levels
1	1
2	2
3	4
4	8
5	16
6	32
7	64
8	128
9	256
10	512
11	1024
12	2048

whereas if you convert the signal from analog to digital logarithmically, these are the levels per f-stop, even at just 10 bits:

F-stop	Levels
1	85
2	85
3	85
4	85
5	85
6	85
7	85
8	85
9	85
10	85
11	85
12	85

10-bit log has more shades per f-stop than the first 7 stops of 12-bit linear.

Its inefficiency is only in the first f-stop, as I understand it, where it is pretty noisy. So you just encoded 85 shades of noise. I think this is an okay trade-off, since the rest of the f-stops aren't starved for bits, or drowned in them.

I wonder if that's why we have Sony S-Log, Panalog, Log-C, etc. These are all slightly different curves. If they all were purely logarithmic, it would be silly to name it after your brand. Just call it log. But Sony, Panasonic, Canon, and Arri have tweaked the curve for the most information-to-data ratio. So I wonder if it's particularly in that noisy low end that they made it a bit more linear, so as not to devote quite so many bits to sheer noise.

Actually, those are the numbers only if your image is purely monochromatic. A 10-bit image is 10 bits per channel, so it's 85x85x85 shades per f-stop, or 614,125. A 10-bit RGB or YUV image has a palette of a little over a billion colors.

Um, no. By very definition a 4:2:2 signal throws away half of the color resolution.

Well, I was referring to SMPTE 292M, which they call uncompressed 10-bit 4:2:2 HD. Sure, the 4:2:2 means that they've tossed half the color resolution away. But usually they distinguish it by calling it subsampling, instead of compression. Compression implies decompression, but there's no getting that color information back. But then again, if you call that compression then every camera is compressed, because right off the bat it has subsampled reality to 4K or 6K or 8K. Any camera whose pixels are larger than the Planck Length would be compressing, by that definition.

It all boils down to what's visible. And anyone who walks by this forum should laugh at us or something, because cameras have been visually good enough since the 5D Mark II.

Yaaaawwwwnnnnnnn............

Yes, TubEfingers has the right idea.

But here's something more practical: $300 vs $60. That's the difference between recording 4K RAW and AVC-Intra 400. RAW, uncompressed 4K is about 1 TB per hour, which is around $300 at the average price per GB --- probably closer to $500. AVC-Intra 400 is about 180 GB per hour, 80% less!

Now actually, if you're going to talk sheer visibility, pure practicality, I think you would be better off setting your GH5 to HD resolution and recording AVC-Intra 200. That is a codec that professional technicians have said is indistinguishable from uncompressed. And we could have a looooong debate about 4K vs 2K again, but let me just say you know which side I fall on. So if you record HD AVC Intra 200, that's actually a codec that's twice as mild as AVC-Intra 400 (5:1 vs 10:1). It's basically the bitrate of ProRes HQ but with a better algorithm. It's only 90 GB per hour, so it has the added ergonomic benefit of making it easy to remember how many minutes a card will give you. Just move the decimal. A card that is 128 GB will give you about 1.28 hours, a 256 GB card will give you about 2.56 hours, etc. Each card will give you a bit more than that, actually, since it's 90 not 100 GB per hour. But just that little ergonomic convenience I think is weightier than the difference between uncompressed RAW and AVC-Intra.

Here's another thing AVC has going for it. Suppose you have the choice between (A) uncompressed, RAW, 12-bit 1920x1080, and (B) 10-bit AVC-Intra 4:2:2 at 1920x1080. Which will look better? Well, if you allow that the AVC-Intra could have been downsampled from a higher resolution (2880x1620 or above), then it can be free of moire and aliasing. Meanwhile the RAW format will either (a) be prone to moire and aliasing, or (b) have an optical low-pass filter that lowers its effective resolution to more like 1400x800. And yet the AVC-Intra is a fraction of the disk space (money). Whatever resolution of uncompressed RAW you trot out, there will always be a lightly compressed doppelganger that is about a tenth the space and visually indistinguishable, even after intensive grading.

I'm not against RAW. I like it for one particular reason: it is the easiest to implement. I would love for another Ikonoskop or Digital Bolex to come out, and if they can't afford the hardware to compress, then just copy that sensor to disk (RAW).

 

[bunk]

I have been following this thread with great interest.

It turned into a discussion between RAW and other codecs. I don’t think the codecs are the culprit. It’s the in camera processing before the codec is written.

Here is a link to the test I did
https://youtu.be/RWa6AK31RD4

I don’t think you can talk about compression without being allowed to cuddle the clips to death. So that's what I did so on all clips equaly.
All GH5S prores files were written to an Atamos but show the exact same problem the GH5 h.624 file has. If I remember correctly the EVA1 files were shot in LongGOP 150 and transcoded to Prores HQ and the 24 p looks fine to me.

The clips were dropped in their native format on a 1080 timeline, meaning everything bigger is cropped. No noise reduction to any of the clips.
First run ungraded, second run graded, third run graded 1quantify node, last run graded 2 quantify nodes.
With the exception of the BMPCC clips, all were collected on line at some point in time.

Alexa 1080 24p Prores HQ 422
Terra 4K 2160 25p prores proxy 422
Terra 6K 1620 23,976p prores HQ 422
EVA1 2160 240p prores HQ 422
EVA1 2160 23,976p prores HQ 422
GH5S 2160 25p prores 422
GH5S 2160 25p prores 422
GH5 2160 24p H.264
BMPCC 1080 24p prores HQ 422
GH5S 2160 23.976p H.264
BMPCC 1080 24p DNG
Axiom 1080 24p DNxHD

The Alexa is king and it shows. The footage just quantifies. No break up, no flicker, no nothing. The terra 6K doesn’t look too good as it seems to suffer from FPN not sure if that’s fixed by now. The Terra 4K proxy clip was added to show how well it would hold up to some of the other camera’s.
EVA1 @23.976 and Axiom -in an early beta state- look great as well. The BMPCC is very close the previous two. No flicker, no added specks to make up for overly strong applied noise reduction (by some referred to as ‘Alien Sightings’), no dancing mosquito pixel clouds, no nothing.
…And then there is the GH5/GH5S well judge for yourself. I personally think it has all of the above…

I was disappointed by the amount pixel clouds my GH4 produced. It turned out to be different but no better on the GH5 and I pointed it out on EOSHD. The owner didn’t coop very well with that message.
In the meantime nothing has changed. To me there is a reason some of us perceive the footage of the GHxx as ‘not filmic’ ‘brittle’ ‘video-ish’ and this quantifying test shows imo pretty well what the reason for this is.
You can get a long way with noise reduction, but alas that will not work on the ‘Alien Sightings’. Stephan Fleet run into them while creating the ‘GH5 firmware 2.0 - Green Screen Tests!!!’ and around 11:00 he makes it sound like he will do some test in the future to find out what is causing this. So there is some hope for the GH5 users.

In the meantime I’ll stick to my BMCC 2.5 mft and BMPCC.

 

[TheDingo]

Here is a link to the test I did https://youtu.be/RWa6AK31RD4

All of the jiggly hand-held footage does not lend itself to comparison with steady tripod mounted footage. Same goes for the EVA1 high-speed footage where the camera can't process individual frames at high speed with the same degree of fidelity as lower frame rates, so it's not really useful for comparison.

The shots made by Andrew Reid of EOSHD fame were shot with beta firmware before the GH5 camera was released, I also think Andrew's night-footage is not useful for comparison.

I know you don't have the resources, but I have much more faith in camera tests where all cameras are filming the same scene under the same lighting at the same ISO setting. The actual differences between camera abilities become quite obvious when tested this way, otherwise you really have no idea what you are looking at. Beta firmware ? An operator that may not know how to optimize the settings for each camera ? Completely different lighting and subject matter ? Jiggly poor camera motion ? Etc...

I plan on buying a BMPCC 4K body to add to my GH5 kit, but I won't trust any camera tests until I see something where all cameras are tested under the same conditions with optimal camera settings.

I remember watching a camera test in 2017 made by the two guys that run the CRFTSHO YouTube channel, where they shot the same scene under the same lighting with the GH5 and the RED Dragon, and in the test they did not identify which camera shot which clip. After watching this test on a 55 inch HDR Samsung TV, it was still quite difficult to tell which camera had shot which clip, the results from both cameras were extremely close.

 

[Matt Gottshalk]

I remember watching a camera test in 2017 made by the two guys that run the CRFTSHO YouTube channel, where they shot the same scene under the same lighting with the GH5 and the RED Dragon, and in the test they did not identify which camera shot which clip. After watching this test on a 55 inch HDR Samsung TV, it was still quite difficult to tell which camera had shot which clip, the results from both cameras were extremely close.

Yeah that clip and my own tests with the GH5 are what convinced me, a former Red Epic owner, to get one too.

 

[sgttom]

Regardless of the consistency of those tests they are pretty extreme. I think there is no doubt Prores HQ is going to beat the GH5 codec. It just comes down to in how much of a difference does it make in practical situations.

 

[joe12south]

I want to share something from a shoot last week.

I had the need to rent the Atomos Shogun Flame (wonderful piece of kit, btw, except for absolutely chewing through batteries!) last week because ProRes delivery was a requirement. For safety, I also recorded to an internal card.

UHD, 24P, 10bit, V-LOG L:
- ProRes HQ
- h.264 400mbps ALL-I

The shot was a slow pan, and I needed to hold the highlights on a very high dynamic range scene, so basically I had to ETTR for the corner of Penn Station. I consider this a challenging scenario, but also a fairly common real world scenario.

Two things were done in post:
1. Rec709 LUT applied (my custom tweak of Panasonic's LUT)
2. Shadows lifted (Curve correction)

ProRes HQ
http://blog.josephmoore.name//galleries/prores_gh5/grab_prores.png

h.264 400mbps ALL-I
http://blog.josephmoore.name//galleries/prores_gh5/grab_h264.png

Notice how ProRes HQ survives the grade with almost no visible artifacts, while h264 reveals subtle macroblock color errors in the sky. Now, these would not have been noticeable had I not had to push the image. h264 compression basically counts on what you record being what you intend to deliver. If you push it out of that zone, you risk revealing where it has used heavier compression. ProRes is much more even handed.

What I found surprising is how well the h264 file held onto shadow detail. I really expected to see more artifacts there, as I did lift the shadows considerably. What should surprise no one is what a beast the GH5 is for resolving detail! I didn't even notice that plane while recording. I have to roto it out of the shot.

 

[TheDingo]

I needed to hold the highlights on a very high dynamic range scene, so basically I had to ETTR for the corner of Penn Station. I consider this a challenging scenario, but also a fairly common real world scenario.

ETTR is NOT meant for high dynamic range shots, so it is the worst thing you could do if you are trying to hold the widest dynamic range your camera can record.

ETTR is meant for more limited dynamic range shots where you can afford to push up the highlights close to the clipping point (but not clipping) which will over expose the shadows to give you as much shadow detail as possible, and then you bring everything down to the proper exposure in post by using the reference chart or grey card target you shot at the beginning. This gives you normal highlights (provided you did not clip), normal mid-tones provided you adjusted the gamma curve to match where these should be, and extremely clean shadows once the overexposed shadows are brought down to where they should be.

I generally avoid ETTR exposures because I want the full dynamic range that the camera can capture, and I prefer to manually clean up the shadows in post with Noise-Reduction profiles calibrated to my camera settings. ( or I just leave the noise as is, which happens pretty often because slightly noisy shadows don't bother me )

Two things were done in post:
1. Rec709 LUT applied (my custom tweak of Panasonic's LUT)
2. Shadows lifted (Curve correction)

Why on earth would you lift the shadows when they are already over-exposed by the ETTR exposure technique? This makes no sense...

 

[joe12south]

ETTR is NOT meant for high dynamic range shots, so it is the worst thing you could do if you are trying to hold the widest dynamic range your camera can record.

ETTR is meant for more limited dynamic range shots where you can afford to push up the highlights close to the clipping point (but not clipping) which will over expose the shadows to give you as much shadow detail as possible, and then you bring everything down to the proper exposure in post by using the reference chart or grey card target you shot at the beginning. This gives you normal highlights (provided you did not clip), normal mid-tones provided you adjusted the gamma curve to match where these should be, and extremely clean shadows once the overexposed shadows are brought down to where they should be.

That's not what ETTR necessarily means or yields. Totally depends on the scene. Yes, for a more limited DR, you would need to push it all back down, but for a high DR scene, it's a way of protecting highlights. They're both valid reasons to "incorrectly" expose a shot. (Though I personally don't use ETTR any more if the scene fits in the camera's DR.)

I exposed to save the corner of the building, making the scene average under-exposed. If I had metered to the middle, the corner would have been completely blown out. Simple as that. Very common scenario.

Why on earth would you lift the shadows when they are already over-exposed by the ETTR exposure technique? This makes no sense...

Just look at the scene. The shadow areas weren't over exposed, they were slightly under-exposed. The result is a nice fat, natural looking scene.

But you're really missing the point. The point is that the internal codec is more brittle than ProRes.

 

[DeeZiD]

The point is that the internal codec is more brittle than ProRes.

That's a huge difference actually. And the internal H265 is even worse in that regard, but the only option without any crazy sharpening. ARGH...

 

[sgttom]

I want to share something from a shoot last week.

I had the need to rent the Atomos Shogun Flame (wonderful piece of kit, btw, except for absolutely chewing through batteries!) last week because ProRes delivery was a requirement. For safety, I also recorded to an internal card.

UHD, 24P, 10bit, V-LOG L:
- ProRes HQ
- h.264 400mbps ALL-I

The shot was a slow pan, and I needed to hold the highlights on a very high dynamic range scene, so basically I had to ETTR for the corner of Penn Station. I consider this a challenging scenario, but also a fairly common real world scenario.

Two things were done in post:
1. Rec709 LUT applied (my custom tweak of Panasonic's LUT)
2. Shadows lifted (Curve correction)

ProRes HQ
http://blog.josephmoore.name//galleries/prores_gh5/grab_prores.png

h.264 400mbps ALL-I
http://blog.josephmoore.name//galleries/prores_gh5/grab_h264.png

Notice how ProRes HQ survives the grade with almost no visible artifacts, while h264 reveals subtle macroblock color errors in the sky. Now, these would not have been noticeable had I not had to push the image. h264 compression basically counts on what you record being what you intend to deliver. If you push it out of that zone, you risk revealing where it has used heavier compression. ProRes is much more even handed.

What I found surprising is how well the h264 file held onto shadow detail. I really expected to see more artifacts there, as I did lift the shadows considerably. What should surprise no one is what a beast the GH5 is for resolving detail! I didn't even notice that plane while recording. I have to roto it out of the shot.

Very interesting. I would be interesting to see how 1080 compares to 4k in terms of the macroblocking.

 

[Thomas Smet]

I want to share something from a shoot last week.

I had the need to rent the Atomos Shogun Flame (wonderful piece of kit, btw, except for absolutely chewing through batteries!) last week because ProRes delivery was a requirement. For safety, I also recorded to an internal card.

UHD, 24P, 10bit, V-LOG L:
- ProRes HQ
- h.264 400mbps ALL-I

The shot was a slow pan, and I needed to hold the highlights on a very high dynamic range scene, so basically I had to ETTR for the corner of Penn Station. I consider this a challenging scenario, but also a fairly common real world scenario.

Two things were done in post:
1. Rec709 LUT applied (my custom tweak of Panasonic's LUT)
2. Shadows lifted (Curve correction)

ProRes HQ
http://blog.josephmoore.name//galleries/prores_gh5/grab_prores.png

h.264 400mbps ALL-I
http://blog.josephmoore.name//galleries/prores_gh5/grab_h264.png

Notice how ProRes HQ survives the grade with almost no visible artifacts, while h264 reveals subtle macroblock color errors in the sky. Now, these would not have been noticeable had I not had to push the image. h264 compression basically counts on what you record being what you intend to deliver. If you push it out of that zone, you risk revealing where it has used heavier compression. ProRes is much more even handed.

What I found surprising is how well the h264 file held onto shadow detail. I really expected to see more artifacts there, as I did lift the shadows considerably. What should surprise no one is what a beast the GH5 is for resolving detail! I didn't even notice that plane while recording. I have to roto it out of the shot.

Joe thanks for sharing. As far as I know you are now the first person to actually post an example to show us any difference between ProRes and internal 10bit from the GH5. The thing that bothers me in both is the amount of mosquito noise around the bird and other edges on both the ProRes and the internal. Forget chroma sub sampling it is this that hurts accurate keying for FX material. It is this mosquito noise that forces us to choke the matte a bit more. Thats why for FX stuff its better to shoot something like Natural that doesn't need to push the values at all. I use an external BMD capture device via thunderbolt to capture HD uncompressed and it eliminates all of this. This is where I don't consider ProRes all that much more superior to the internal formats. As you have posted in some fine gradient situations ProRes can hold up better but from a VFX perspective both are about the same.

Just to be clear was this image from the middle of the shot or the first frame? The first frame always shows a bit more macro blocking before the encoder can really kick in.

 

[sgttom]

Joe thanks for sharing. As far as I know you are now the first person to actually post an example to show us any difference between ProRes and internal 10bit from the GH5. The thing that bothers me in both is the amount of mosquito noise around the bird and other edges on both the ProRes and the internal. Forget chroma sub sampling it is this that hurts accurate keying for FX material. It is this mosquito noise that forces us to choke the matte a bit more. Thats why for FX stuff its better to shoot something like Natural that doesn't need to push the values at all. I use an external BMD capture device via thunderbolt to capture HD uncompressed and it eliminates all of this. This is where I don't consider ProRes all that much more superior to the internal formats. As you have posted in some fine gradient situations ProRes can hold up better but from a VFX perspective both are about the same.

Just to be clear was this image from the middle of the shot or the first frame? The first frame always shows a bit more macro blocking before the encoder can really kick in.

I wonder if HLG would be superior to Vlog in terms of the mosquito noise

 

[visceralpsyche]

I want to share something from a shoot last week.

I had the need to rent the Atomos Shogun Flame (wonderful piece of kit, btw, except for absolutely chewing through batteries!) last week because ProRes delivery was a requirement. For safety, I also recorded to an internal card.

UHD, 24P, 10bit, V-LOG L:
- ProRes HQ
- h.264 400mbps ALL-I

The shot was a slow pan, and I needed to hold the highlights on a very high dynamic range scene, so basically I had to ETTR for the corner of Penn Station. I consider this a challenging scenario, but also a fairly common real world scenario.

Two things were done in post:
1. Rec709 LUT applied (my custom tweak of Panasonic's LUT)
2. Shadows lifted (Curve correction)

ProRes HQ
http://blog.josephmoore.name//galleries/prores_gh5/grab_prores.png

h.264 400mbps ALL-I
http://blog.josephmoore.name//galleries/prores_gh5/grab_h264.png

Notice how ProRes HQ survives the grade with almost no visible artifacts, while h264 reveals subtle macroblock color errors in the sky. Now, these would not have been noticeable had I not had to push the image. h264 compression basically counts on what you record being what you intend to deliver. If you push it out of that zone, you risk revealing where it has used heavier compression. ProRes is much more even handed.

What I found surprising is how well the h264 file held onto shadow detail. I really expected to see more artifacts there, as I did lift the shadows considerably. What should surprise no one is what a beast the GH5 is for resolving detail! I didn't even notice that plane while recording. I have to roto it out of the shot.

Excellent example of the codec differences Joe

I'm very excited for the BMPCC4K and have one on pre-order.

Cheers,

Paul

 

[combatentropy]

I don’t think the codecs are the culprit. It’s the in camera processing before the codec is written.

I think you may be on to something, and thank you for your examples. Thank you, joe12south, for your comparison too, which is more controlled.

If I zoom all the way in on joe's, yes the ProRes looks better. What bothers me is how it looks better. First of all, the ProRes is twice the bandwidth of the AVC-I, so you would expect it to be better. But the AVC-I doesn't look like just heavier compression. What I mainly see is less noise. JPEG, ProRes, and AVC all use the Discrete Cosine Transform at the heart of their compression. The DCT has many artifacts, as most of you are familiar with --- mainly mosquito noise and blockiness. What the DCT doesn't do is smoothe things out, not like this. I mean, it will smoothe things out if you go really low quality, but by then the edges of the blocks also start appearing. Here the areas of smoothed-out-ness have splotchy outlines.

This leads to seriously wonder if the GH5 first puts the signal through another layer of noise reduction before recording to AVC.

An even more accurate comparison would be, if anyone can, between ProRes and AVC at more similar bitrates. One way to do that is to record both in HD, not 4K. ProRes HQ at HD resolution, and AVC-I 200 are about the same data rate. I don't know why Panny didn't put in a AVC-I 800 for 4K. That would have been nice.

Like I said, JPEG, AVC, and ProRes all are very similar at their core. Sure, they have tweaks here and there, but at the exact same data rate, given the exact same source, I would think their differences are almost indistinguishable, even zoomed all the way in.

 

[joe12south]

Just to be clear was this image from the middle of the shot or the first frame? The first frame always shows a bit more macro blocking before the encoder can really kick in.

Good point. This is midway through the pan, though with the ALL-I version of the CODEC, would that really make a difference? (I'm almost always trimming the beginning, so I've never noticed.)

 

[joe12south]

I think you may be on to something, and thank you for your examples. Thank you, joe12south, for your comparison too, which is more controlled.

If I zoom all the way in on joe's, yes the ProRes looks better. What bothers me is how it looks better. First of all, the ProRes is twice the bandwidth of the AVC-I, so you would expect it to be better. But the AVC-I doesn't look like just heavier compression. What I mainly see is less noise. JPEG, ProRes, and AVC all use the Discrete Cosine Transform at the heart of their compression. The DCT has many artifacts, as most of you are familiar with --- mainly mosquito noise and blockiness. What the DCT doesn't do is smoothe things out, not like this. I mean, it will smoothe things out if you go really low quality, but by then the edges of the blocks also start appearing. Here the areas of smoothed-out-ness have splotchy outlines.

This leads to seriously wonder if the GH5 first puts the signal through another layer of noise reduction before recording to AVC.

An even more accurate comparison would be, if anyone can, between ProRes and AVC at more similar bitrates. One way to do that is to record both in HD, not 4K. ProRes HQ at HD resolution, and AVC-I 200 are about the same data rate. I don't know why Panny didn't put in a AVC-I 800 for 4K. That would have been nice.

Like I said, JPEG, AVC, and ProRes all are very similar at their core. Sure, they have tweaks here and there, but at the exact same data rate, given the exact same source, I would think their differences are almost indistinguishable, even zoomed all the way in.

So, I'm not positive about this because it has been a very long time since I've actually read the ProRes whitepapers, but I think one of the distinctions between it and many other delivery codecs is that it is more even handed. h264, h265, JPEG, etc. rely on weakness in human vision to compress certain parts of the signal more than others (low contrast areas, and dark areas being the most obvious.) I *think* that ProRes does not do this, and that makes the image more malleable. (At the cost of a higher data rate, obviously.)

What we see as "noise", a codec considers high frequency, low contrast data. h264 compresses the hell out of this, more or less acting like crude noise reduction. In the uncorrected V-LOG L file, you'd be hard pressed to see the difference between the two skies. It's only when the contrast is increased when they become apparent. (Which makes sense, the subtle compression errors get exaggerated.)

 

[joe12south]

Joe thanks for sharing. As far as I know you are now the first person to actually post an example to show us any difference between ProRes and internal 10bit from the GH5. The thing that bothers me in both is the amount of mosquito noise around the bird and other edges on both the ProRes and the internal. Forget chroma sub sampling it is this that hurts accurate keying for FX material. It is this mosquito noise that forces us to choke the matte a bit more. Thats why for FX stuff its better to shoot something like Natural that doesn't need to push the values at all. I use an external BMD capture device via thunderbolt to capture HD uncompressed and it eliminates all of this. This is where I don't consider ProRes all that much more superior to the internal formats. As you have posted in some fine gradient situations ProRes can hold up better but from a VFX perspective both are about the same.

The in camera sharpening has to be exacerbating this, too. I shot at -5, and as you can see, there is still artificial ringing. The GH5 is crazy detailed as it is, I REALLY wish sharpening could be turned off.

 

[John Brawley]

Notice how ProRes HQ survives the grade with almost no visible artifacts, while h264 reveals subtle macroblock color errors in the sky. Now, these would not have been noticeable had I not had to push the image. h264 compression basically counts on what you record being what you intend to deliver. If you push it out of that zone, you risk revealing where it has used heavier

Exactly.

As previously mentioned, higher bit depth and more robust codec options will allow you to push the image around further.

For anyone used to shooting reversal film it's the same thing. It looks great when you nail the exposure and you're happy with how it looks out of camera. But if you want to then push it around in the grade you will very quickly find the short comings of more compressed codecs.

As previously mentioned, to talk of codecs being visually indistinguishable doesn't count for much once you start manipulating the image with any post processing. And the keying you do for the grade (not the same as chroma keying for VFX) where you're trying to grab subtle secondary colours will also show up very quickly any shortcoming in the data.

JB

 

[DeeZiD]

The in camera sharpening has to be exacerbating this, too. I shot at -5, and as you can see, there is still artificial ringing. The GH5 is crazy detailed as it is, I REALLY wish sharpening could be turned off.

Are these shots sharpened in post?

 

[Thomas Smet]

Saying all DCT formats should look the same is like saying McDonalds, Wendy's, Culvers and Red Robin all have the same burger because they all grill ground beef and slap it in a bun.

DCT is just a method to convert images to blocks of data. Two formats can absolutely be DCT based and have the same data rate and have a very different look. It has happened in the past (Photojpeg at 50mbps vs DV50 at 50mbps.). Even two different h264 or mpeg2 encoders can result in a different look and quality with the same data rate.

The GH5 formats are odd little beasts and if the share anything with AVCIntra then they do use a form of motion prediction even for the all-I 400mbps format. Thats partially how 400mbps can get closer to the look of ProRes HQ at 800mbps. Half the data will eventually lose something in translation. That loss seems to come in the form of subtle gradients if pushed very hard.

Wit that said the GH5 still is one of the few cameras that attempts to give users an affordable 10bit 4:2:2 format. No other camera under $5000 really does that. Perhaps the GH5 is not a grading beast in some situations but a lot of others have graded GH5 material with little to no problems. There is also the fact that not every videographer or even film maker grades at that level. The GH5 and GH5S still beat the pants off of other 8bit DSLR and video cameras out there when it comes to grading. Plus for the grading fanatics the GH5 camera do have HDMI out which is again something nobody else provides at 10bit besides BMD. Thats why I now consider Panasonic and BMD the only realistic players in the game.