The great 8K debate. Why I have changed my mind

[zlfan]

Many years ago, a pro was complaining on a famous photo forum, that his sony a9 camera (which was one of the top cameras at the time) and zeiss lenses setup was stolen. He was so upset, that he bought a T2i and the kit 18-55 lens. He specialized in photo wall business. He said that for A9 and zeiss lenses, he only needed about 20 photos, for t2i and kit lens, he needed 50 photos. However, he spent more than 10 grand for software stitching these photos. I guess that he also needed the special nodal point tripod head. 

Not sure if anyone experimented with 4 4k cameras with the nodal tripod head to combine into a 8k stream. For still scenes, you just need one 4k camera to move to 4 different positions. But for moving scences, then you need 4 4k cameras recording at the same time, which defeats the purpose of saving money. 

[zlfan]

Sorry, Sony A900 instead of A9.

[Matt Kieley]

I would probably only use high resolutions (4K and up) for projects with a heavy amount of VFX which is where I would find the ability to scale and reframe useful. Or if I specifically want that super crispy look. In term of visual quality FHD and 2K are more than adequate for narrative films. Most movie theaters are only projecting 2K DCP. 35mm film prints were only about as sharp as 720p. I even kinda prefer a softer look, and I suspect a lot of others do considering how many people use high res raw shooting cinema cameras, and put vintage lenses and diffusion filters on them. Many of my favorite films are shot on 16mm, and I've always loved that soft, grainy look. 

I'm a big fan of/greatly respect David Fincher, who uses the highest resolution cameras he can for more flexibility. I personally don't like shooting this way. I prefer to have less flexibility in post. I get optional paralysis and would be too tempted to keep tinkering and tweaking endlessly when the options are limitless. It's easer to force my choices in production because the options and time are limited enough to allow me to make some snap decisions, or to force other decisions. I'd rather get the look I want in-camera.

I'm more interested in seeing how upscaling software will advance. The martial arts action film The Raid, which was shot on the Panasonic AF100 (remember that camera??) is being upscaled/remastered in 4K and the frame grab samples released by the director look good.

 

[markr041]

More 8K frame grabs from RAW video:

 

[kye]

3 hours ago, markr041 said:

More 8K frame grabs from RAW video:

 

There's a bit more noise in these images than I would have thought.  Were you keeping to base ISO?

It is 8K, so by the time its used in the real world then this will have mostly cleared up, but by that logic there's no point in having 8K - might as well have had better/bigger pixels.

[markr041]

6 hours ago, kye said:

There's a bit more noise in these images than I would have thought.  Were you keeping to base ISO?

It is 8K, so by the time its used in the real world then this will have mostly cleared up, but by that logic there's no point in having 8K - might as well have had better/bigger pixels.

Yes, always base ISO (800). And I do not think that clip was underexposed overall and pushed up in post. Could be just as you say, small pixels necessary for high resolution.

[Michael S]

I was always a bit surprised that, while in the audio world everyone is familiar with the Nyquist theorem and how obviously you must sample with at least twice the sample frequency as the frequencies you want to reproduce, and a bit of spare so you can make a well-behaved low-pass filter, in the world of video this doesn't seem so obvious. I blame in part the manufacturers who loved to put a 1920x1080 bayer sensor in a camera and then to advertised it as being an HD camera while in the audio realm you wouldn't be able to cheat with sampling frequancies. Of course for video much more bandwith is needed so right from the start the engineers tried every trick they could think of to reduce bandwith needs.

So yes, having an 8k bayer sensor, with an OLPF to avoid aliasing would provide an excellent source for a 4K video stream. If the recording is 8K that means you could also manipulate the image without immediately running into scaling artefacts when preparing a clean, detailed 4k delivery.

8K delivery? Useful for specific use cases where people have their faces pressed up to the screen or where the image is supposed to fill your entire field of view as with VR but otherwise not very useful.

 

[Django]

Z9 FW4.0 has a ISO200 mode (Lo 2.0) that dramatically cleans up noise in N-Log:

Not sure why Z8 hasn't yet received this update (among other features)..

[ND64]

8 hours ago, Michael S said:

having an 8k bayer sensor, with an OLPF to avoid aliasing would provide an excellent source for a 4K video

8k bayer is 8k. It doesn't need math, you only need to see with your own eye. Modern demosaicing algorithms with the help of AI are more advanced than you think.

[kye]

11 hours ago, markr041 said:

Yes, always base ISO (800). And I do not think that clip was underexposed overall and pushed up in post. Could be just as you say, small pixels necessary for high resolution.

800 isn't really base ISO for the sensor, just for that mode, and as @Django mentioned they could do one at ISO 200 that is a lot cleaner.

I'm fine with that level of noise, but that's because I'm a fan of cinema which doesn't require anything even remotely close to 8K - it's really just for cropped modes and pixel peeing folks..

8 hours ago, Michael S said:

I was always a bit surprised that, while in the audio world everyone is familiar with the Nyquist theorem and how obviously you must sample with at least twice the sample frequency as the frequencies you want to reproduce, and a bit of spare so you can make a well-behaved low-pass filter, in the world of video this doesn't seem so obvious. I blame in part the manufacturers who loved to put a 1920x1080 bayer sensor in a camera and then to advertised it as being an HD camera while in the audio realm you wouldn't be able to cheat with sampling frequancies. Of course for video much more bandwith is needed so right from the start the engineers tried every trick they could think of to reduce bandwith needs.

So yes, having an 8k bayer sensor, with an OLPF to avoid aliasing would provide an excellent source for a 4K video stream. If the recording is 8K that means you could also manipulate the image without immediately running into scaling artefacts when preparing a clean, detailed 4k delivery.

8K delivery? Useful for specific use cases where people have their faces pressed up to the screen or where the image is supposed to fill your entire field of view as with VR but otherwise not very useful.

I think people are broadly aware of oversampling and its advantages.  However the other thing to keep in mind is that you don't need to double the nyquist frequency - you only need to have a slight advantage, like audio being recorded at 48K and then delivered at 44.1K, or cameras like the GH5 which was 5.2K downsampling to 4K.

28 minutes ago, Eric Calabros said:

8k bayer is 8k. It doesn't need math, you only need to see with your own eye. Modern demosaicing algorithms with the help of AI are more advanced than you think.

The more I learn about what is going on under the hood, the more I realise that statements like "8k bayer is 8k" don't even make sense.  To truly un-pack that statement would require a whole textbook, and that's just the technical side and ignoring the perceptual aspects!

[kye]

8K is cancelled.

Just use AI to make it better...  IMAX does!

https://ymcinema.com/2023/10/02/imax-ceo-we-use-ai-to-blowup-images/?expand_article=1

From IMAX CEO, Richard Gelfond:

• "we use it to blow up images. We use AI to make the images look better, we sharpen the edges, and we take the grain out. We have been using AI for supplements for a while."
• "However, the best reference for that utilization of the IMAX proprietary algorithm and AI tech, is the talked-about sci-fi project, The Creator. The movie was shot entirely on Sony FX3 which is not an IMAX-certified camera. Nevertheless, the ProRes RAW footage was undergone special treatment by IMAX AI technologies, in order to boost the imagery and make it capable enough to hold up against the huge canvas."

Other streaming services only store feature films in 2K and upscale to 4K for people who stream in 4K, and now IMAX, the supposed best quality folks upscale using AI.

We all knew that streaming was a low-quality distribution, and now IMAX is too...

How can they get away with this????  

Maybe.... *gasp* ....because it's not visible?  I mean, you could say that AI is good enough for IMAX, but you could also say that visual perception is so low that you can't even tell that IMAX is upscaling with AI!  It works both ways!!

😂😂😂