Calling 'fake' on the iPhone computational photography glitch | Hacker News Reader

Calling 'fake' on the iPhone computational photography glitch

daringfireball.net

66 pointsyowzadave37 comments

Both Gruber's "fake" explanation and the "computational photography" explanation are likely wrong.

There's a third explanation: It's almost certainly an accidental use of panorama mode.

- It's two poses, not three, with the stitch somewhere on her back.

- The shop attendant was taking the photos. Coates wouldn't know if panorama mode was accidentally selected for one.

- The Photos app info pane doesn't flag that a photo was taken in panorama mode. (But you can infer it from the nonstandard resolution if you know of this possibility.)

- The top of the right mirror is very slightly curved, indicating the perspective correction of panorama mode. (Not the railing, which is likely curved in real life, but the actual mirror frame.) It's barely visible to the naked eye, but you can verify by drawing a straight-edge with a lasso selection tool.

I replicated it at home with two mirrors (technically, one mirror and one iPad with the front-facing camera) and my iPhone in panorama mode, and a kitchen timer for "proof" (demonstrates over 2 seconds apart). You can barely see the panorama stitch and wouldn't notice it unless you were looking for it: https://mastodon.social/@jeff_tyrrill/111509692643373685

> top of the right mirror is very slightly curved, indicating the perspective correction of panorama mode.

Straight lines are actually always curved in wide angle photos (you can tell because of the exaggeratedly stretched corners) unless you perfectly correct for it. It's called radial distortion and is a natural property of lens projection. This is not on its own a sign of stitching, just imperfect calibration.

Anyway, I also believe this was a pano shot, but a curve isn't real evidence of that.

I don’t understand why people seem to „want“ this to be real. It obviously isn’t. The effect cannot be explained by rolling shutter and therefore would involve actively stitching together multiple shots. This would be huge feature that Apple would advertise (take a family photo and you get a version where everyone smiles). Apple does not a/b test stuff like that without telling users, that’s just not the way their hype machine department works.

The way rolling shutter and computation photography works that kind of glitch is exactly the kind that is possible

Not common by any means but plausible

Edit: tried it with my iPhone and it’s definitely tricky to do, if I get a photo that demonstrates this effect I’ll post it publicly. The window of opportunity to take such a photo is extremely tiny

And it would definitely be much easier to photoshop it than to capture it “in camera”

Rolling shutter won’t cause this: the readout of the whole frame happens in a fraction of a second, and I don’t see someone moving their arms quickly enough to trigger this kind of glitch. You usually see rolling shutter effects with objects that are moving very fast, like wheels and propellers.

Yeah for sure, it would take a freak accident to move your arms so quickly and suddenly and have them end up sharp in the end result.

The closest I’ve been able to achieve in a few minutes with my phone is a blurry vs non blurry image of my hand in the reflection vs in real life. Pretty hard to pull off this kind of double exposure trick when the software isn’t actually intending to take double exposures

I wonder if this is easier to pull off with "night mode"? A few tests suggests that the iPhone is indeed quite choosy with the exposures that it stacks, so there's definitely a possibility of it picking-and-choosing parts of pictures to stack.

And perfectly stable hand what held that phone, not making a blur anywhere.

Optical image stabilization can explain that part

Is it possible that the phone was very busy/lagging and what would normally take a fraction of a second took a couple of seconds?

The camera hardware implements the readout upon command from the application processor (what you'd normally call the "CPU" of a phone), and the readout does have to be carefully timed in order to ensure a consistent exposure across the frame. So, whether the phone is overloaded or not will not affect readout speed (and therefore the rolling shutter effect).

This does not preclude the possibility that some other photographic process was stalled, such as HDR stacking.

Perhaps this was some kind of rare memory fault where it took a series of photos and there was some kind of corruption that caused elements of each to be merged.

Or perhaps there was some weird fault that meant different sections of the photo were taken at discreetly different times and then stitched together.

Rolling shutter doesn't produce such clean discrete artifacts, but more of how rotors are bent for helicopters when readout between motion isn't fast enough. This image is as if global shutter took 3 separate shots of 3 separate poses within a space of 0.12~0.15s each

Yeah my bad this isn’t rolling shutter, this is multi frame compositing and likely filtering out blurry frames randomly because the arms did move

I just tried a little personal experiment and I can manipulate my arms into all 3 positions well within 0.5 seconds, without exaggerating the movement. Start with your hands clasped (as in the right most image), then put your arms down but starting with your left. It also seems plenty bright enough to capture without motion blur. It could hardly take long to do some basic zonal stitching too. Having a hard time following why the case should seem so definitive. iOS does facial recognition for focusing ... wouldn't surprise me at all if the computational photography took a close-in-time separate still of each person, to get them as in focus as possible. It'd be pretty cool.

If this is what happened, why the story about going to the apple store (+wizard stuff???)…

Have you never been given some bs excuse by someone?

Or maybe he said "it could be" and that got left out of the quote.

Update. It was a panorama: https://www.youtube.com/watch?v=QqcYiDn6d1s

The petapixel article with the photo in question says an Apple tech said Apple is beta-testing a feature that chooses the best picture from several taken. Does Apple roll out beta features to a subset of users like that?

> Does Apple roll out beta features to a subset of users like that?

No; it's explicitly opt-in, you have to go out of your way to install beta versions of iOS.

You used to have to sign up and install a profile. As of iOS 15, it’s a toggle in the update settings. It still takes a deliberate action to enable though.

Probably need to have iOS beta enabled explicitly or have a TestFlight account to actually access such feature. I too feel this photograph was stage managed somehow

> I too feel this photograph was stage managed somehow

The example does feel “too clean”.

Has anyone at all been able to reproduce something similar?

To have this kind of perfect reproduction, 3 poses need to be in 3 separate global shutter readouts (at best) within space of 0.12~0.15s i.e a burst at about 8-10 fps.

iPhone sensor still raster-reads so I would have expected more motion between these three poses in that time span. I feel by the pose and image quality, there was more time in between

Apple also basically never just drops new features without marketing them somewhere. They would have said that best take was coming to iOS 17.1 or something, then rolled out those beta versions with that feature.

What would be her motivation to lie about this?

Attention? Don't underestimate people's ability to come up with stupid stuff for attention.

It could be to get attention.

She runs a podcast called “Nobody Panic”. The blurb for it says “Your guidebook to being a fully functioning adult without screaming all the time. Each week, Stevie Martin and Tessa Coates tackle life's big, small, fun and sometimes scary questions with the help of experts and special guests because we're all in this together, guys.”

So yeah, why would she have any motivation?

She’s not necessarily lying but she’s definitely an accomplished story teller who knows how to use something to a promotional advantage.

> Claude Zeins takes me up on my challenge.

I think Gruber meant to link that sentence to this post: https://mastodon.social/@czeins/111508306096972808

So in conclusion, she might be lying but she probably isn't, and she's most likely using iOS beta?

What a weird thing to get upset about. I'm not sure what is the point of this post...

Gruber is an Apple apologist / one-man stealth marketing department. Voicing his view on topics like these is literally his thing.

He’s not “apologising” for anything here. Physics and a rudimentary understanding of computational photography should be triggering sceptical responses like this. As someone who has been taking photos using analogue and digital SLRs for over 35 years, while playing with exposing and editing photos using analogue and digital tools, this image definitely looks suspect. It is too sharp and too posed.

You’ve been banging the freedom and choice drum on the original thread, and while I don’t necessarily disagree with some of the points you made there, his scepticism jibes with your hot take and I think it would be fair to describe you as anti-Apple, so there’s that.

Based on what we know about Apple, physics and ML as applied to photography, the likelihood is that this is a staged photo. I’m more surprised that petapixel didn’t do a more thoughtful evaluation rather than perpetuating the story.

Reply on news.ycombinator.com