Alpha Beta Omega
Veteran
There's not only bitrate or codec that makes a good image. Signal processing is just as important and with a small power efficient camera like the GH5 there's just so much that can be done.
GH4 'Lab Review Reveals 10bit Flaws in Vlog + FREE GH5 LUT'
- Thread starter Bern Caughey
- Start date
morgan_moore
Major Contributor
I used to be a 10bit fanatic. But when doing a test I got the exact opposite result that I expected.. what kills an image? 10bit or high compression.. I found the latter much more damaging.
So what should be the expectations of 150mbs 4k? .. well it is less data than 50mbs HD, in fact it is not enough data to hit the BBC minimum.. and it breaks... quel suprise
My compression test video showing banding or not under a hard grade..
So what should be the expectations of 150mbs 4k? .. well it is less data than 50mbs HD, in fact it is not enough data to hit the BBC minimum.. and it breaks... quel suprise
My compression test video showing banding or not under a hard grade..
Thomas Smet
Veteran
H264 is not just a DCT based encoder however. There is a lot of other stuff going on that helps it handle images. Besides we are not talking equal bitrates here either. 400mbps is extremely low compared to 800 with ProRes HQ. ProRes also has different levels of quality such as Regular ProRes and ProRes LT. Each level drops the data rate used and the image starts to fall apart. LT saves some space but you can start to get some blockiness in the image. H264 all-I at 400mbps is right around ProRes LT which is 410 for 30p. That may be fine for some people, especially if they down convert to HD but ProRes LT can also display a lot of artifacts.
Another thing to keep in mind is no two encoders are the same. Back when we still created DVD's for clients we had a lot of different Mpeg2 encoders. They all encoded at 8.0mbps and yet there was a huge quality difference between all of them. A lot of Mpeg2 encoders also implemented multiple passes to help get the quality at a better level for a given bitrate. You can't do multiple passes in a camera since the encoding is live. A lot of these encoders were also optimized at different levels for either speed or quality. An encoder has to do a lot of work and it cuts corners to make sure it can encode faster. In the case of a camera it has to encode in realtime so 150mbps H264 in a camera may not be as visually accurate as 150mbps H264 encoding on a computer which can takes its time and optimize how the bits are divided.
So no 800mbps H264 is not the same as 800mbps ProRes. Two encoders handling how it builds the DCT blocks in very different ways. 800mbps from one H264 encoder is also not the same as 800mbps from another H264 encoder.
XDCAM HD422 is another example of a well optimized mpeg2 format compared to other mpeg2 encoders also encoding 50mbps. Both are mpeg2 4:2:2 and use 50mbps but can deliver very different results. Just taking any mpeg2 encoder and encoding 50mbps 4:2:2 mpeg2 is not the same as XDCAM HD422. The same is true when I developed mpeg2 encoding tools for the Avid Liquid software. I created an extension for the editing program that opened up 300mbps 4:2:2 HD mpeg2 encoding. On paper this should have been equal to Prores422 HQ but visually it was not. It looked great but it was still based on a specific mpeg2 encoder and mpeg2 encoding practices even though it was an all-I format. Still beat the pants off of HDV at that time but there were different artifacts compared to other formats like ProRes or DnxHD.
Thomas Smet
Veteran
Agreed. Even recording externally 8bit ProRes422 from the GH4 shows significant improvement with V-log. There is still some potential for banding but most of the odd color artifacts go away. 10bit helps but it is the H264 encoder in the GH4/5 that is causing most of the damage to V-log. I still always use 10bit because every little extra helps.
32bit float grading tools really help a lot but if the damage is already done that damage is there forever. If an image looks solid with 8bit then a 32bit float grading tool should keep it solid even when pushed hard. We shoot 10bit to make sure that source is as solid as it can be no matter the circumstances. The benefits will not always be obvious but they matter when needed.
I think you're confusing bit depth and dynamic range here.
Ergonomics. Attaching an external recorder will always add size and weight, which is not always desirable or practical.
That's a bit of a contradiction there. Click bait refers to a misleading headline leading to content that has no value.
I'm glad they did, and here's why.
I bought the GH4 in no small part because of all the favorable reviews with the overall image quality. And my first tests with V-Log on the GH4 were favorable. But as I looked closer, I saw the flaws others were talking about, flaws which made the footage unusable. My editor saw the same. I then did the in-depth research I should have done before buying the GH4 and ended up switched to the BMPCC. Had I done that research ahead of time, I might now own two usable BMPCCs instead of one usable BMPCC and one GH4 that gets left behind.
I only hope these guys will do these same tests again once the 400 Mb/s mode drops in August. For myself, 400Mb/s I-frame, 10 bit, 4:2:2 in V-Log is the only shooting mode that interests me. If that cannot at least approach the image quality of the BMPCC in RAW mode, I will likely not switch back to Panasonic.
So I do appreciate these nitpicky tests, having been burned before by not heeding them until after purchase.
Last edited:
The other way to look at it is that All-I generally requires roughly twice the data rate of IPB encoding, so the 400 Mb/s IPB equivalent would be 200 Mb/s. That's 50 Mb/s more than the current 150 Mb/s. It could make a worthwhile difference.
Barry_Green
Moderator
Depends on the encoder chip and what toolset it has available to it. h.264 has many "levels" or "profiles", and not all h.264 encodings use all the available tools. As Thomas points out, you cannot get multi-pass encoding on a camera, because all processing is done live, so that's one toolset that simply is not available to the GH5. Frankly, I don't expect that a tiny little camera running off a tiny little battery is going to have the horsepower available to it to be able to employ the highest levels of h.264 encoding.
There's so much more to it than just bitrate. I guarantee you that a studio mastering box employing all the tools available in the highest profiles will churn out h.264 footage that beats the holy crap out of the h.264 that the GH1 could output, at half or less the bitrate that the GH1 used. The GH1, being the earliest camera in the lineup, used the least advanced h.264 chip, and its h.264 encoding was simply terrible -- the worst recording format I think that has ever been released on any camera ever, even much worse than the original HDV.
It may be possible that the state of h.264 hardware chips has advanced to the point where a powerful chip is now small enough and power-efficient-enough to fit in a tiny camera like this and be powered off a tiny battery like that, but I don't know. All I'm saying is -- ProRes is a mature technology and a known quantity, we know exactly what we will get out of it. We don't know what the profile or level of the GH5's 400mbps All-I will be, but based on the experience we have with the All-I 200 HD, I (and others) are not terribly optimistic that 400mbps is going to prove sufficient for encoding ProRes-level quality UHD at 10-bit and 4:2:2.
Depends on the profile of h.264.
Yes, but only in certain profiles. We don't know what profiles the GH5 chipset supports. So just because it's technically possible at some level, does not mean that it will be implemented in a chip that meets the size and power draw requirements to be part of a small DSLM camera.
The technology isn't, no. But the fact that most H.264 encodings are 8-bit 4:2:0 interframe with low bitrates is probably more representative of the state of hardware encoding chips that can be fit in, priced for, and powered by, a small camera. Unless there's been a quantum shift in h.264 chip power, or pricing, then the new chip is probably going to be comparable to what we've been seeing in the last few generations of the camera. It'll be improved, I have no doubt, but the level of improvement will probably be on par with the generational level of improvement from the GH1 to the GH2, from the GH2 to the GH3, and from the GH3 to the GH4. Will that level of improvement get it to the point where it can deliver ProRes-like results in half the bitrate? That seems extraordinarily optimistic for an onboard chip in an inexpensive camera powered by a tiny battery.
filmguy123
Veteran
So the moral of the story is, pray for an Atomos Inferno feature level recorder at NAB 2017, but in a near Small 502HD / iPhone 7/7 Plus form factor - small, lightweight, rugged, 800mbps ProRes 240fps 1080 and 4K 60p.
filmguy123
Veteran
So how do you feel about 400mbps All-I for 10bit 4:2:2 1080p HD?
Thomas Smet
Veteran
There is no science or fact behind the twice the data rate thing. That is just a rough assumption and even then there are a lot of other factors involved. For the most part 100mbps on the GH4 looks very good but its those details that typically are not seen that get tossed out. The claim of twice the data rate is for visual quality and not pixel perfect quality. Once you toss out detail you can never really get it back 100%. Just like with 4:2:0, once you toss out the color samples you can try to estimate the missing values but it will never be 100% accurate. What H264 does is try to match the overall visual look but details are still missing. Details that can pop up when grading or pulling a chroma key. The same is true with all-I. All-I just means it doesn't use the same type of motion prediction to fake the in between frames. A lot of the other data reduction methods are still there. IPB can have an extra amount of macro blocking when motion gets high but all those other data saving factors are still in place.
So no IPB will not be 100% the same as all-I at half the bitrate. Visually to the naked eye it may appear to but there is a lot of tossed out information there.
Thomas Smet
Veteran
I'm not sure there is a moral here. Yes internal 10bit will have some loss of data but I think 99% of the people will be hard pressed to actually notice it. The color issues are seriously splitting hairs and even then only really show up if you use V-log. The article even states the other profiles look very good. What is in question here is really is H264 even at 10bits still not good enough for V-log. I think the answer is it really depends. For most people who thought 4k on the GH4 looked amazing it will look even better on the GH5 and V-log is finally usable internal. With the Gh4 you could use V-log internally but not without some artifacts. With the Gh5 most of those are reduced to a point where they will not be noticed as much. So the 10bit internal is still a great format for those who really want V-log and don't want to use an external recorder. We have seen plenty of great samples from film makers like Luke Neumann that show us just how good it can look. It is much better than internal V-log on the GH4 so a huge improvement.
4k 60p ProRes422 HQ needs 1,768mbps. Unless you absolutely have to have 4k 60p ProRes there are plenty of external recorders out there that can handle up to 30p 4k. Video Devices even makes a 5" screen if you really want compact and small. Personally I use an Atomos Blade to record external HD at 24p from my GH4. I don't really need 4k and just want as clean of a HD video as I can get. It is a 5" screen but I kind of now prefer a 7" screen. 5" is still too small to really help with focus so if you can handle the tiny bit of extra weight and bulk I say get a 7".
Cary Knoop
Veteran
And you say that ProRes does not do that?
If the general thought here is that ProRes has some magic under the covers that other encoders with similar bitrates and settings don't have then no sorry I don't buy that at all.
ProRes is DCT based just like H.264 and uses macroblocks as well.
If you want something fundamentally different look for instance at the Cineform codec.
Barry_Green
Moderator
We'd have to see it. That's plenty of bitrate, depending on how the codec uses it. AVC-Intra looks very good for 1080p 10-bit 4:2:2, and it's only 100-mbit All-I. AVC-Ultra is supreme, and it's 200mbps All-I. But those codecs have a lot of horsepower behind them. 400mbps in the GH5 is enough bandwidth to make pristine 1080p, but it all depends on how efficient the codec is and how much advantage it can take of that 400mbps.
Cary Knoop
Veteran
I found the 30p Perpetua Dolly shot extremely noisy.
I will definitely buy a GH5 because it is such a practical camera with a good performance similar or slightly better than the GH4.
But so far the new 10bit feature has disappointed me and I personally don't see how we can pass all blame on this to H.264