Samsung caught faking zoom photos of the Moon
theverge.com
theverge.com
>Earlier this week, we were made aware of Samsung’s Game Optimizing Service (GOS) and how it throttles the performance of games and applications. GOS decides to throttle (or not to throttle) applications using application identifiers and not application behavior. We view this as a form of benchmark manipulation as major benchmark applications, including Geekbench, are not throttled by this service.
That's why the only valid GPU reviews are video reviews you can watch, not metrics.
I wouldn't say 'caught' or 'in trouble'. They're applying some AI blackbox to their photo tech, and the lack of transparency is annoying people so then people make viral posts like this claiming they 'caught' Samsung (when in reality it reads like an AD for Samsung).
Is it actually “physics defying” or just can’t be done with available technology?
There is no known technology that would make it physically possible to fit such a camera in a mobile phone, just as there is no known technology that would make it physically possible to travel faster than light.
But yes, the manufacturers are now squeezing the last few percent out of the diffraction limit rather than leaving most of it on the table, so the year-by-year leaps and bounds in camera quality are mostly coming from AI enhancement not optics or sensors.
Images are formed by interference of plane waves propagating at different angles. To resolve finer details, your optical system needs to collect waves that were propagating away from the source (the moon) at larger angles. This means you can either get closer to the moon or increase the lens diameter. It doesn't matter how the lens works, you could use any photonics magic trick you want, you're still limited by the (aperture) size of the optical system.
Samsung's lens is purportedly too small to resolve the level of detail produced in the moon images.
The physical properties of diffraction establish fundamental minimum limits on the requisite lens aperture diameter for a given resolution.
See: https://en.m.wikipedia.org/wiki/Angular_resolution#The_Rayle...
Also: https://en.m.wikipedia.org/wiki/Dawes%27_limit (simplified, for the more common use case)
>> Samsung’s process is more intrusive than this: it doesn’t just improve the sharpness of blurry details — it creates them.
Same as if they used the same trick with a model Coke can, and you tried to take a blurry photo a of an insect on your can.
With the moon, it's especially ridiculous because no one wants Samsung's prefab moon photo, we have tons of those. They want to see the phone zoom in on what they are actually seeing, just for fun. I don't know, maybe if you photograph the moon behind a tree, you'd like a more detailed moon. But it's Moon "Zoom" , not Space Zoom, and it's Virtual Zoom. It's like the "Just Egg" vegan egg, but Just Egg knows you know it's not a real Egg. Samsung just keeps lying and pretending we're stupid, instead of advertising the nifty feature for what it is. That's tacky.
Samsung's argument is "if you can't tell, does it matter"?
This is new post-truth AI world.
A smart move, actually, since most people only care that their phone takes decent selfies at the Instagram resolution. Heck, I don't even know what each of the three cameras at the back of my new phone is for. I stopped keeping track of those megapixels once they went above 10MP or so.
I wonder what people really expect though, afaik the camera has 10x optical zoom and after that up to 100x digital zoom. What does digital zoom mean? I'd say using normal scaling and other algorithms to improve quality.
I'm sure professional photographers already feel that way, as they often don't take photos taken with a phone serious as photography.
Showing pretty pictures of the moon on Instagram? Doesn't matter.
Planning a rover trip across the surface of the moon? Does matter.
And while it may be absurd to expect anyone is planning the latter, the problem is that nowhere in the marketing or documentation of the future as-exists is there sufficient information to determine what use cases it should not be used for.
Or, in other words, if Samsung were doing the exact same thing for cancer detection microscopy, they'd be killing people...
Now that's some major whataboutism if I've ever seen one.
Of course, if they use this tech for cancer detection, I will happily join the protests.
But they aren't. They take pictures you take with your camera and make them look sharp. If anything it's super cool the way they do it without resorting to pre-saved pictures of the moon. It's literally absurd to me that people are getting annoyed at this, even though literally all phones do image processing of one kind or another.
>>the problem is that nowhere in the marketing or documentation of the future as-exists is there sufficient information to determine what use cases it should not be used for.
Really? That's the problem people are having with it? That the mode exclusively demonstrated and marketed for taking pictures of the moon doesn't have sufficient information to establish what use cases it should not be used for? I find that extremely hard to believe - it sounds like a made up problem for clicks, exactly what Verge is doing(they are kind of famous for it).
You're right (I assume) it is an ML thing and not "replacing" the image, but it takes a 170x170px intentionally blurry blob and turns it into a detailed moon photo. Very very little of that photo is the result of light captured via the camera sensor.
This is the original posters side-by-side:
" Very very little of that photo is the result of light captured via the camera sensor."
The photo is entirely the result of the light captured by the sensor - after image processing, yes detail is added which physically couldn't have been captured by the camera - but like I said above, that's literally all mobile photography nowadays. I literally don't see the problem, if anything I'm extremely impressed by the result.
[1] https://news.ycombinator.com/item?id=35107601 [2] https://news.ycombinator.com/item?id=35123389