GH4 Can someone explain the 4:2:2 output from GH4?

[Joshua Cadmium]

Okay, while the bucket analogy seems apt, it's really not going to work in the digital realm, where data (or gasoline) can be created at will out of thin air. On top of that it ignores the whole science of statistics. Modern society wouldn't exist without interpolating data based on limited sampling.

Again, I'm not saying that a Bayer sensor is going to give you perfect per pixel color like you'd get with a Foveon sensor - just that it will be significantly (you'll be able to easily see it) better.

I realize I did get the Arri information wrong - I thought there was an option for 2880 x 1620 (or higher) ProRes, but the only options are in fact a downres. However, Arri does claim that "luminance resolution ... correlates to the sensor pixel count." - http://www.arri.com/camera/alexa/workflow/working_with_arriraw/arriraw/de_bayering/. Plus you've got the Amira upres, but let's not open more cans of worms.

Rather than debating each argument point by point, I think a visual example would help. Here's the difference between the Blackmagic Pocket in 4:2:2 ProRes HQ and raw: http://vimeo.com/79231766 . According to the arguments presented so far, there shouldn't be that much of a difference between the two, but the actual difference is massive - including color fidelity.

While the compressed 4:2:0 out of the GH4 is okay, the 4:2:2 color I've seen is fantastic and more than a significant step up from the current in camera recording. Raw video would be a massive step up.

 

[harddrive]

Okay, while the bucket analogy seems apt, it's really not going to work in the digital realm, where data (or gasoline) can be created at will out of thin air. On top of that it ignores the whole science of statistics. Modern society wouldn't exist without interpolating data based on limited sampling.

We're not arguing against the need for interpolation. Whatever the source, if you have a 4:2:0 final signal, you'll need to interpolate to get something suitable for viewing - the display is likely to want R,G,B for every pixel. The only question is where in the chain to do any interpolation.

And whilst "guessed" data may be able to be created out of thin air, that's not the same as real data.

If I gave you a sequence with every other number missing, say, 1,x,5,y,7,z,13 - what data would you create for the values of x,y,z? I'd be surprised to get any answer other than x=3, y=6, z=10. So you may decide to transmit the "4:4:4" sequence of 1,3,5,6,7,10,13 - I'd transmit the "4:2:0" sequence of 1,5,7,13.

But in fact the original sequence was 1,2,5,5,7,12,13, so what you send carries no more "real" data than what I do - whilst being more cumbersome to send. In the video world, that equates to either higher bitrate or higher compression.

I realize I did get the Arri information wrong - I thought there was an option for 2880 x 1620 (or higher) ProRes, but the only options are in fact a downres. However, Arri does claim that "luminance resolution ... correlates to the sensor pixel count." - [/quote] I think you are reading f...es HQ and raw: [URL]http://vimeo.com/79231766 .

Hmm. Not sure how much that proves, and it doesn't really cover or specify the full workflow. I would never argue that there are advantages to recording RAW (and disadvantages as well) though they are more seen to be an ability to put off "burning in" a final look. I tend to see this example as more saying "with RAW I can regrade in post better than with coded video" - and yes, that can only be true. But for chroma resolution? I think the example I posted above with the coloured resolution charts showing how a given sensor is far better with a black-white image than a coloured one visually shows the issue far more precisely.

While the compressed 4:2:0 out of the GH4 is okay, the 4:2:2 color I've seen is fantastic and more than a significant step up from the current in camera recording. Raw video would be a massive step up.

But is this talking about 1080 or 4k? If 1080, then yes, I can see how that could be true. As for RAW, then I doubt it would NECESSARILY give any better quality per se - what it certainly would do is keep more grading options open, without doubt. But at the expense of a longer workflow - not good if you're shooting for a quickish turn round.

 

[Thomas Smet]

Okay, while the bucket analogy seems apt, it's really not going to work in the digital realm, where data (or gasoline) can be created at will out of thin air. On top of that it ignores the whole science of statistics. Modern society wouldn't exist without interpolating data based on limited sampling.

Again, I'm not saying that a Bayer sensor is going to give you perfect per pixel color like you'd get with a Foveon sensor - just that it will be significantly (you'll be able to easily see it) better.

I realize I did get the Arri information wrong - I thought there was an option for 2880 x 1620 (or higher) ProRes, but the only options are in fact a downres. However, Arri does claim that "luminance resolution ... correlates to the sensor pixel count." - http://www.arri.com/camera/alexa/workflow/working_with_arriraw/arriraw/de_bayering/. Plus you've got the Amira upres, but let's not open more cans of worms.

Rather than debating each argument point by point, I think a visual example would help. Here's the difference between the Blackmagic Pocket in 4:2:2 ProRes HQ and raw: http://vimeo.com/79231766 . According to the arguments presented so far, there shouldn't be that much of a difference between the two, but the actual difference is massive - including color fidelity.

While the compressed 4:2:0 out of the GH4 is okay, the 4:2:2 color I've seen is fantastic and more than a significant step up from the current in camera recording. Raw video would be a massive step up.

You can't compare a baked in ProRes look from a BM camera and the RAW output and say the 4:4:4 vs 4:2:2 is the reason for the quality difference here. The color sampling has about the least amount of effect on what you are seeing here. Image compression formats even ProRes throw away data. ProRes is not uncompressed. ProRes also has to have a baked in look with on the fly debayering applied while it is recording. There are about a million different things going on here that affect the quality. Not just the color sampling.

4:2:2 output from the GH4 looks better because it isn't compressed. H264 while a stunning format really butchers color at 100mbits. Not just because of the 4:2:0 but because of macro blocking and other forms of DCT image compression.

I also don't think there is a complete lack of quality difference between a bayer sensor and a 4:2:0 recording. Some of us have already said bayer may be a bit more refined because of how it offsets the blue and red 2x2 pixel blocks. There is a difference yes but it is nowhere near significant and subtle at best. Most of what you are seeing are the benefits of RAW, 12bit, 0% image compression and 0 in camera processing.

In the end color sub sampling does not affect overall image quality by a lot. It doesn't magically make the image look more pristine or look like a higher production value. This is a myth mainly from the fact that most cameras that could record 4:2:2 or RAW are typically significantly better then a camera that can only record 4:2:0. Yes there is a reason why a RAW camera tends to cost more than a DSLR and it isn't just the ability to shoot 4:4:4. Most of what you are seeing taking place in typical 4:2:0 formats is the H264 butchering of the image and very little to do with the 4:2:0 part of it. The poor neglected color is hammered by a heck of a lot more than the color sub sampling. Lets forget about color sampling theory here for a second and keep in mind we are talking about massively compressed images.

Ever wonder why so many "pros" would debate heavily on if a new cameras HDMI output was 4:2:0 or 4:2:2? No one could really tell without massive pixel peeping at 400 or 800%. What does that tell you? That in the end the color sampling matters the least. It is getting rid of that image compression that is the bigger concern. When you strip away all other factors that affect image quality and truly measure the pure difference between 4:2:0 and 4:4:4 you start to see it isn't as significant after all.

Of course if you can get better go for it but by no means should it be considered a quality deal breaker.