Hacker News Reader

Comment by rwalle | Hacker News Reader

While I understand what the author tries to say, I have to point out that ship has long sailed. Samsung just pushed it a bit too far and slapped a "scene optimizer" label on it.

AI has been used in "cell phone photography" for a few years, at lease since Pixel 2 where a mediocre sensor produced much better pictures than what people expect (maybe there are other players who did this even earlier). And every manufacturer started doing it, including Apple. Otherwise, do you think "night mode" is just pure magic? Of course not, algorithms are used everywhere.

How do you define "fake"? In podcasts, Verge editor Nilay Patel has asked various people "what is a photo", because the concept of a "photo" has become increasingly blurry. That is the question the author is asking, and people may have different answers from the author's.

View on HN

> Otherwise, do you think "night mode" is just pure magic?

Night mode definitely uses some AI but most of the result is from stacking frames. Samsung here did not label it as a "scene optimizer". Their marketing just calls it Space Zoom. The only disclaimer they provide is "Space Zoom includes digital zoom, which may cause some image deterioration."

According to Samsung -- and I just confirmed it on my S22 under Camera -> Camera Setting -- it's called "Scene Optimizer"[1]:

  [ Overview of moon photography]

  Since the Galaxy S10, AI technology has been applied to the camera so that users can take the best photos regardless of time and place.

  To this end, we have developed the Scene Optimizer function, which helps AI to recognize the subject to be photographed and derive the optimal result.

  From Galaxy S21, even when you take a picture of the moon, ai recognizes the target as the moon through learned data, and multi-frame synthesis and deep learning-based ai technology when shooting. The detail improvement engine function that makes the picture clearer has been applied.

  Users who want photos as they are without AI technology can disable the optimum shooting function for each scene.

[1] https://r1-community-samsung-com.translate.goog/t5/camcyclop...

Note that this behaviour is limited to scene mode, which has a moon shoot mode. You can always use the normal or pro mode where the pictures are not magically enhanced.

Is is ridiculous that OP consider this "cheating". Most people just want a nice picture and don't give a damn about AI.

Photo is an interesting word. It's meaning is clarified by other words, such as photorealism, photofinish. These words will (strictly) lose their meaning if photograph simply means image captured and processed by a device.

Curiously and revealingly, the political word photo-op stands alone in this photo- parade of words in the age of photo-imaginings. The universe does indeed have a sense of humor.

Using algorithms to take multiple pictures and stack them together is fine. The information is real, exists, and objective. People in the background won't (for example) suddenly be facing the other way because of the algorithm.

The problem is that AI isn't just interpolating data. It is wholesale adding extra data that simply doesn't exist. The person in the background is facing left, but the sensor couldn't possibly have captured that detail even after multiple images--it was a coin flip that the AI made.

The issue is that, like privacy, most people won't care ... until they do. By that time, it will be too late.

> People in the background won't (for example) suddenly be facing the other way because of the algorithm.

Someone here included an example where it does do something like this: https://news.ycombinator.com/item?id=35109568

The software technology in the original pixel cameras were using multiple frames of varying exposure to allow for impressive dynamic range in images while still retaining colour and contrast. This is quite a difficult thing to do as requires precise understanding of what the 'edge' of an object is, and I think that is what AI was used for. This stacking technique is also used for night exposures.

I'm sure that they have started using AI to fill in details more recently, but this is just to point out clever use of multiple exposures and AI can help without faking detail.

eg "Do We See Through a Microscope?" https://philpapers.org/archive/HACDWS.pdf

> the concept of a "photo" has become increasingly blurry

nice.

Reply on news.ycombinator.com