Apple Reportedly Reduced Face ID Accuracy to Speed Up iPhone X Production

[Braderunner]

Everyone should avoid iPhone X until next year.

 

[Jsameds]

Perhaps, but what you wrote was inaccurate. The article clearly states the accuracy level was reduced. You posted:


Everyone is talking about it, because the article's title, and contents mention the accuracy was reduced.

With zero context though. It's redundant information without context, was it reduced by 50% or 0.0005%?

 

[BootsWalking]

Yes, but even for an unknown source, you should get an indication of where the source is. Supply chain? Distribution? Apple themselves? This article is just vague rumours.

Oh, and take a look here:

https://www.bloomberg.com/search?query=apple

Not one single positive article on the first page of results. That should tell you something.

They gave an indication of the source - they said someone close to the matter. Any more specific might reveal the source who doesn't want to be revealed.

The first page of your link represents only two days of articles. I didn't realize they were required to publish positive articles about Apple every day.

 

[IG88]

Do you really think that Apple would announce something and than not hold the promise? IF there are it is less acurate, which is not proven yet, as it is only a report, Apple would have made the decision before the presentation. They can not put up a flashy keynote slide that says 1:1'000'000 and not deliver that. That would be a liability which users could sue over.

What group has the ability to prove the accuracy isn’t 1,000,000:1? How do you even quantify that?

 

[Glideslope]

I'm interested to see just how many fewer Dots are projected now. Seems the Dot projector has been the bugaboo since day one. Did they go from 30,000 to 25,000? Did they go from 30,000 to 15,000? I think this is critical to know before spending $1,200.

 

[maflynn]

A conspiracy theorist would say that another big company is behind those "leaks". But yeah, it seems obvious that somebody wants this phone to be in a bad light.

I don't think Apple needs any help to put this phone in a bad light, they'

With zero context though. It's redundant information without context, was it reduced by 50% or 0.0005%?

Agreed, we really have no idea if this is an issue or not.

 

[VulchR]

I've been mildly amused by your posts in this thread. You've made some fundamental misconceptions regarding Face ID, the tech behind it, and biometric identification in general. Wasn't gonna say anything but you addressed me, so...
Here's an example of what I'm talking about. Your claim that Apple was eerily silent about false negatives is a complaint that doesn't stand scrutiny. You've essentially introduced a red herring to further a narrative. An incorrect one at that.

Apple clearly addressed false negatives. They did it in the most embarrassing way possible - a failed demo - but they addressed false negatives nonetheless. You simply enter your passcode. Just sayin'...

Yes you can enter in a passcode, but I prefer to avoid that hassle, particularity on an expensive phone. The question is how often will you will have to do that, particularly if you wear glasses. So far as I know Apple has not released that information. We all know the false positive rate cited in the key note. So, if you know the false negative rate from Apple literature please give the number and cite the source.

Also please point out the fundamental misunderstandings I have about the system, but be aware I have studied the visual system as a neuroscientist, I work with people who study face perception, I am familiar with neural network modelling (including knowing many of the people who first worked on back-propagation and created the first neural nets for speech synthesis and speech recognition), and I understand a certain amount of signal detection theory (which applies in this case).

 

[SwissGuy93]

Perhaps, but what you wrote was inaccurate. The article clearly states the accuracy level was reduced. You posted:


Everyone is talking about it, because the article's title, and contents mention the accuracy was reduced.

Not really. The title is "Inside Apple's Struggle To Get The iPhone X To Market", yet everyone reporting uses the headline that the accuracy is reduced. And Bloomberg is technically speaking also contradicting itself. They say that Apple reduces the accuracy of Face ID. But then again they only talk about the dot projector and have that statement at the end:

"It’s not clear how much the new specs will reduce the technology’s efficacy."

So, they already scream that Face ID's accuracy is reduced, yet they do not even know if it will be noticeable at all. That is just clickbait journalism in my eyes.

 

[DNichter]

I question how valid it is, but I wouldn't worry too much about it either way. As long as it works well, it really doesn't matter to me. If we get our new X and Face ID is flaky, then yea there will be a lot of unhappy people. Apple isn't that dumb and definitely wouldn't allow anything that would negatively affect usage of the new feature.

 

[nfilg]

X means - You can't buy it!

 

[MacGizmo]

Sounds like compromising the quality of a $1000 product to me... Insane how Apple choose to do this in a rush to meet demand instead of rolling out quality units at a slower pace.

Insane how people take everything published on a blog as bible.

 

[Ntombi]

Erm... in addition to any concern about coatings on lenses blocking light, which does not appear to be the case, there is the issue that lenses alter the path of light, and hence the inference of the 3D shape of the head as illuminated by the laser dots. You can practically see around my head with the lenses I wear.... but sometimes I also wear contacts. I cannot imagine the system will be able to handle both contexts, and hence my concern that unlocking the phone will be a hassle.

You mean like the demo? *cough*

I have been using Apple products since the Lisa (yes, I have actually word processed on a Lisa), so it's not as though I am an Apple naysayer. I would imagine FaceID will work for the majority of people the most of the time. That still doesn't mean that the computational problems posed by things like glasses, or even puffiness of the face from allergies, won't throw the process off. We'll see how well Apple's neural networks can fill in missing information if parts of the face are obscured etc. However, going from TouchID (2D image under closely controlled lightening conditions of a body part that does not change) to FaceID (3D image with potential portions obscured by glasses etc.) is magnitudes more difficult.

We'll see how well Apple solved almost insurmountable problems. My guess is that they haven't and so when the phone is sold there will be something that says that FaceID will work only under certain conditions. As I have noted above, I have not seen any demonstration of FaceID on somebody wearing glasses and then contacts, and we don't know whether FaceID declines in accuracy if somebody wears glasses.

Pretty much every question you are posing has been answered already. You can keep being annoyed and repeating the same thing, but we aren’t pulling the answers out of thin air.

Apple posted a white paper about Face ID, and MacRumors posted a précis. Allergy puffiness, glasses, beard scruff, are all accounted for.

 

[phpmaven]

"Reduced accuracy". If true, it's not really what you want to hear from a $1000+ gadget. How about you reduce your prices along with it, Apple!

The story says that they "relaxed" a specification. Assuming this story is even true, which is a big assumption, it doesn’t claim that accuracy is reduced. All this really means, again if true, is the tolerances of one the components didn’t need to be quite as tight as they thought to make everything work properly. This is a non-story. Of course all of the usual suspects here will freak out.

 

[Chupa Chupa]

Everyone should avoid iPhone X until next year.

At least the 256GB GSM model. At least wait until about 3:30am eastern on Friday. They should have some of the kinks out of that model by then.

 

[VulchR]

Yes of course infrared radiation bounces of frames. But that is why you have 30'000 dots. So that enough dots will match your data in the neural engine. And please just go to the website and you see that the model there first wears glasses then not, then has a beard then not and so on. And the glasses will for sure not be part of your profile so different glasses will not pose a problem.
Also; There is not one person that wears glasses all the time. So if Apple says that glasses are supported then the assume that people get that it means that you can wear them or not...

OK, finally making progress.

No doubt you are correct in that the neural net makes a de facto template to which the pattern of dots serving as the input are compared. We also know from the keynote that the template is dynamically updated (probably each time you use FaceID there are incremental changes to the template). It can also be assumed that the template captures spatial correlations among the infra-red dots, so can handle partial obscuring of the face. The question is how the template matching process is affected by variability with respect to the user wearing one or more pairs of glasses or contacts. As it stands now, we don't know. That is all I am saying. And I note so far as I can tell nobody seems to be able to find an actual statistic in the Apple literature that specifies what the false positive rate is. Any engineer working on this problem would use signal detection theory to describe how well the system works. Apple has the information, they just haven't released it. Again, that makes me have a healthy scepticism.

 

[5105973]

Someone needs to email Craig. Though he is software I’m curious if he responds. Doesn’t seem like Apple, but who knows?

While Craig is articulate and charismatic and seems approachable, he is not their PR person and not part of the supply chain, so he would not be the appropriate person to ask. And please don’t distract him from overseeing all the bug fixes we need.

 

[newyorksole]

Sounds like compromising the quality of a $1000 product to me... Insane how Apple choose to do this in a rush to meet demand instead of rolling out quality units at a slower pace.

The problem isn’t Apple. It’s people believing every single thing they read. Just because an article is published doesn’t make it true. Until I use Face ID for myself I will take this with an entire grain of salt.

 

[KPOM]

A company reduces the accuracy of their key feature just to be able to up their production? I don't believe this for a second. It would be utterly stupid if they did this.

It’s still 20 Times as accurate as Touch ID. Maybe they were aiming for 25 times. Put it into perspective.

 

[jarman92]

If this is true it’s a huge disappointment, but unfortunately not a surprise for Apple lately. They should have just delayed the release by a few weeks if they’re having so many issues...not like it would be the first time. And their reputation will take way more of a hit if FaceID quality is poor vs the iPhone X is late.

That said, it seems unlikely Apple would reduce standards if they’re still going to end up with only 2-3 million units on launch day...

 

[Bazooka-joe]

The Apple website says that Face ID is enabled by the TrueDepth camera and is simple to set up. It projects and analyzes more than 30,000 invisible dots to create a precise depth map of your face.

Is that what it said before.

 

[VulchR]

Pretty much every question you are posing has been answered already. You can keep being annoyed and repeating the same thing, but we aren’t pulling the answers out of thin air.

Apple posted a white paper about Face ID, and MacRumors posted a précis. Allergy puffiness, glasses, beard scruff, are all accounted for.

No full statistics about recognition performance are given in the white paper (only false positives mentioned), just claims. Moreover, the paper discusses gradual adaptation of the FaceID as your face changes. I understand that. I could probably even deliver a lecture about how that is likely to occur because it relates to some of the scientific research I conduct. However, what happens is there are sudden changes to the appearance of the face (e.g., taking on and off glasses, wearing different kinds of glasses)? Does that confuse the neural net analysing the pattern of dots? That is all I am asking, Apple has not released information about that. So, no, my questions have not been answered by Apple.

 

[KPOM]

I call BS on this article, but if it is true then all the more reason to get one on launch day as supposedly the earlier ones would have more accurate FaceID.

They may have reduced it from 97% accuracy to 95% accuracy, I can live with that. But as I said before, I think it's BS just to get clicks.

If they made a decision it would have been during development, not final production.

 

[5105973]

Wow, if true, that's a bummer.
First if the sensor is so fragile, will it go out of spec during normal wear and tear of the phone?

Secondly, Apple seems to keep making questionable choices, like adding FaceID when no one asked for it, there was no need and its inferior to TouchID. It may be superior in time, but it seems the technology is not ready for primetime yet.

From everything I’ve read and seen demonstrated in leaks and the official public demos, the technology works as it is supposed to. It’s good to go...there is just the question of whether or not it can be made in the quantities and timeframe Apple needs to meet projected demand and desired sales numbers. They sell such an unimaginable number of phones. I guess we could build an impressive LEGO style mansion out of iPhone boxes for this release alone!

 

[jcmeyer5]

There is absolutely no need for this technology, other than Apple trying to leverage the hardware they know will be in the phone to reduce components (and costs).

 

[Steve121178]

I am usually one who gets up in the middle of the night and pre-orders. This is is just another reason I will wait until the "S" version of this phone so I can get the version that has had all the kinks worked out.

I will definitely wait for the 2018 version. It's too much money to spend on something that will clearly be heavily refined, optimised & improved in less than a year.