Hacker News Reader

Comment by pjc50 | Hacker News Reader

Gloomy prediction: based on Samsung Moongate, the next logical step after the smartphone camera replacing the DSLR is AI replacing the smartphone camera. Or at least substituting for underlying camera quality. Take a 10MP image and have the AI fill in the detail. Or remove detail (see the google advert for editing people out of your holiday photos). Or put in some detail that wasn't there.

View on HN

Smartphone photos are already heavily ML-processed, with actual details and colors being replaced by what the model thinks looks best. The moon thing was just the most blatant example.

I haven’t noticed any invented details or colors in smartphone photos that I’ve taken myself. Most of the claims of this that I’ve seen online are not very convincing. Of course you sometimes get artifacts from sharpening and de-Bayering, but those can occur with any digital camera.

No kind of camera sees "actual colors", and neither do your eyes; the actual colors are the same day or night but that doesn't mean you can see them at night.

When there is light, the colors you see depend on how well you're adapted to its white point.

new from Samsung. it's a black box perfectly square just to make Jobs turn over in his grave. there are no lenses, but it's the most incredible camera you've ever seen. it has a microphone where you tell it the image you want...

"I want to see a picture with me at the Grand Canyon with nobody else in the shot at sunset after storm has just passed with a rainbow in the background"

waits a few seconds, boom. post to social, get lots of envious comments about my cool vacation.

I can't decide if in the future we will be bloblike sloths in chairs like in Wall-E, or maybe it will be like The Matrix but you are plugged into a peleton bike.

where did you get the peleton reference from The Matrix? even the machines realized that was wasted. you spend your entire life cycle in a sensory deprivation tank with the computer telling you how much of a good time you are having. wasting energy on actually moving muscles is absurd! that's not thinking like a machine

I figured the peloton would be the method of entry for the machine. Start off with the rich people doing it for working out to make it seem sexy and cool (done), get everyone else on it (in progress), get your boss to buy pelotons for remote zoom calls (probably has happened), start buying energy from peleton users pedaling (will be pitched soon), offer drug to allow sleeping while pedaling and making energy, change formulation to prevent consciousness entirely while pedaling and remain pedaling until your knees explode and you are replaced on the bike by the machines with another drugged up biological generator.

Watch the episode called 15 Million Merits from Black Mirror.

While I agree that ML-based image-enhancement and photography/videography assist at capture time will continue to offer highly useful capabilities, I don't agree when that argument is used to support the separate contention that mobile imaging will make larger format ILC/DSLR imaging irrelevant. The reason is that ML (and computational photography in general) is fundamental tech that can improve all categories of imaging. It's a metaphorical tide that lifts all boats equally from pro battleships and tankers to hobbyist racing sloops and yachts to consumer speedboats and jet skis.

(note: I'm setting aside the current misguided one-click "make it better" AI features as a temporary aberration that the marketplace of consumer tastes will correct. There's already backlash emerging around over-enhanced AI images on social media.)

This is already the case with cameras in consumer phones.

"subscribe to android photos plus to have all the ai applied billboards removed from your photos"

Too bad we can't get the opposite: a DSLR controlled by an AI that manipulates the camera parameters just as a human would, or an AI driven interface, that captures what you're seeing without an AI hallucinating or ML touching up the photo. Panasonic can't even get autofocus right.

Check out this AI camera control module: https://witharsenal.com/

It's not something I'd use, I usually want as much creative control as possible over all the settings and RAW conversion. There are too many permutations of parameters that "correctly" capture the scene according to different aesthetic aims. The reviews of the version 2 seem to agree, that it's useful, but not for experienced photographers with their own creative vision. But how then would the inexperienced ever develop a personal vision, using this?

Oh, nice, that's pretty cool

Aren’t things like the focus stacking and better HDR already part of MagicLantern? The new things seem to be the deep colour and the crowd removal.

Reply on news.ycombinator.com