The first use of 4k wasn't for 4k delivery, but for single camera shoot of talking head reframing. Full frame for the wide, cropped in for the MED/CU shots. Larger image also allows for cleaner image stabilization. Allows for more detail on pulling keys in post, etc.
There are lots of benefits from larger image than delivery.
I'm not sure this is the case. Low-ISO cinema 35mm film (like Kodak VisionT 50D/250D) far exceeds SD resolution. How do you think Star Wars was ever scanned into 4k? Or 2001: A Space Odyssey? Film captures highlight detail logarithmically as well, while digital sensors still haven't really figured out how to render highlights well. This allows greater shadow detail with a skilled cinematographer with no loss of quality in brightly lit scenes. Theoretically, a single 35mm still can be scanned into 4k with no loss in detail, and higher-resolution scans won't lose quality -- there's just dimishing returns and in the end you might just be getting higher-res grain, but it is a far cry from standard definition.
Even 16mm can do HD scans just fine.
There's a reason Tarantino, Wes Anderson, PTA, Nolan, Sam Raimi, Scorsese, Spielberg, and David Lynch still swear by film (even though many of these directors have dabbled with digital). The idea that film is inherently "low-res" is a really frustrating and common misconception among younger generations unfamiliar with the technology. I find it kind of unbelievable that DPs would shoot 35mm just for an SD scan. That type of use case seems better suited for mini-DV or other contemporary video formats.
>That type of use case seems better suited for mini-DV or other contemporary video formats.
You realize that people were shooting film for TV production long long before DV was ever invented right?
Lots of shows were filmed on 35mm. For example, Star Trek: The Original Series and Star Trek: The Next Generation were filmed on 35mm, so in modern times they've been able to re-scan them and make HD Blu-ray releases. I believe shooting direct to video for anything higher budget than daytime soap operas didn't become commonplace until the mid-90's.
This probably was also to deal with period TV cameras. Video cameras of that time couldn’t deal with bright lights and had terrible dynamic range. CCDs improved this a lot. Direct to video productions looked cheap.
They shot 35mm for the big screen.
They weren't doing for screen size considerations. It's just what they had. You shot 35/16 or you shot video (shudder).
What GP is talking about is the antialiasing low-pass filter used when downsampling video to smaller resolutions, or when initially sampling it at the sensor. This is the same exact concept as the antialiasing low-pass filter used when downsampling audio to lower sample rates, or when initially sampling it at the ADC.
When they don't, like the Canon 5D mark ii when used in video mode (which just threw out most of the sensor's pixels to get to 1080p), you get aliasing producing moire patterns like this:
Font hinting is an example of a nonlinear downscaling though most people don’t think of it that way. It preserves line sharpness, which is a non-bandlimited feature since lines are infinitely sharp = made of infinite frequencies.
Overly sharp linear filters produce artifacts like blockiness and ringing; nonlinear ones would try to avoid that but you could still get more difficult ones like moiré.
This is good, because it means we can finally treat computer graphics as properly scalable images without introducing "blurriness".
Note that lines were never infinitely sharp; pixels aren't infinitely sharp, they have a shape and the screen construction determines how sharp they look. They are, however, sharper than what you'd get in a nyquist-limited system.
Even if you don't do any other post-processing, downscaled video looks better than video captured at sensor resolution. This is because you can implement a steeper low-pass filter (closer to ideal brick-wall filter) in software than you can in actual optics. Hardware lowpass filters tuned to block most stuff above nyquist will also block a lot below it.
A 1080p video captured off a good 1080p sensor will have vastly lower amplitude at, say, 500 line pair vertical than a 1080p video captured on a 4k sensor and lowpassed in software with a good filter.