Apple Says Claims Face ID Was Downgraded to Speed Up Production Are 'Completely False'

[IG88]

Have you ever trained a neural net? It doesn't work that way.

It feels like you’re just tossing around technical jargon with “trained a neural net.”

My company runs clash detection on 3D point clouds on a weekly basis. From what I’ve gathered, that’s exactly what the red dot projector and infrared camera are creating - a 3D point cloud. Repeatability is far far far more important than the raw number of dots.

Assuming that Apple relaxed a tolerance, the raw number of dots output could be reduced just fine as long as the dot projector repeatability still falls within the required precision.

 

[Abazigal]

https://www.aboveavalon.com/notes/2017/10/26/apple-is-facing-a-double-standard

And I am inclined to agree with the writer. For some people, Apple just can’t seem to do anything right, and their every move is closely scrutinised and held up to impossible standards, while the competitors are praised for moves which later turn out to be flops. Moved which were initially used as examples that Apple was falling behind.

It’s getting ridiculous.

 

[bizack]

It feels like you’re just tossing around technical jargon with “trained a neural net.”

My company runs clash detection on 3D point clouds on a weekly basis. From what I’ve gathered, that’s exactly what the red dot projector and infrared camera are creating - a 3D point cloud. Repeatability is far far far more important than the raw number of dots.

Assuming that Apple relaxed a tolerance, the raw number of dots output could be reduced just fine as long as the dot projector repeatability still falls within the required precision.

I'm not actually - I have an MS in Comp Sci and studied machine learning. The 3D mapping of your face from the dot projector uses neural nets and machine learning. The reason so many images of faces were used by Apple to train the neural net that's used for processing images is to distinguish between animate/inanimate images, variations of the same face (i.e. with glasses, a hat, facial hair, etc), differentiating between a face and the background, etc. That's what training a neural net involves. You don't just throw some arbitrary number of points at a face and then determine 'yes, that's so and so' and then decide to change the parameters of the network (number of points projected would be one of those parameters).

 

[lazard]

https://www.aboveavalon.com/notes/2017/10/26/apple-is-facing-a-double-standard

And I am inclined to agree with the writer. For some people, Apple just can’t seem to do anything right, and their every move is closely scrutinised and held up to impossible standards, while the competitors are praised for moves which later turn out to be flops. Moved which were initially used as examples that Apple was falling behind.

It’s getting ridiculous.

When you're the leader of the pack (and boast about it), you're held to a higher standard.

 

[NT1440]

And you know what I was getting at, too. Companies distort the truth for salsa and marketing all the time. They (like politicians) choose their words very carefully. And often a claim that sounds like one thing means, when you carefully review the language or details, something else.

My point, in response to the other poster, is that holding up Apple as some beacon of righteousness and truth is silly. They're a company, and as such they say and do things in the best interests of the bottom line. And as I stated in my previous post, Apple's carefully worded "rebuttal" is not NECESSARILY in conflict with the content of the Bloomberg story. That's all.

I don't hold apple up as a beacon of righteousness, if this where a fair world they'd have their asses handed to them for the absurd amounts of tax evasion scheming they participate in and potentially given the corporate death penalty. Just because I admire the top notch engineering they do doesn't mean I'm an apologist.

In this particular regard, given that every single year in the period leading up to an iPhone launch there are dozens of just flat out falsehoods from media outlets (likely fed information by stock manipulators) I don't lend any credibility to reports that can't be bothered to include any technical information. The fact that most of the tech "journalists" these days do little more than demos and reviews doesn't help much.

If you see a pattern from the same people year after year, you tend to start paying attention to it. Do yourself a favor and look at the last 5 iPhone launches. Without exception every single year the same "supply chain" analysts and "sources with knowledge inside Apple"come out of the woodwork in the weeks leading up to launch to tell everyone how thing A has changed, or somehow the manufacturing process is still in flux mere weeks before millions are to be sold. It's pure ****ing nonsense to even entertain the notion that hardware for a product that has to be manufactured in the millions is somehow still in flux less than 6 months before production ramps. That's lunacy.

 

[Abazigal]

When you're the leader of the pack (and boast about it), you're held to a higher standard.

And I am all right with that.

I just ask that the rest of the competition be held to the same standard as well. Why is a product like face ID (or even Touch ID back then) treated with so much skepticism even before it is released, while other products like the google pixel praised to the high heavens, even as problems abound?

Even funnier are the people who claim that Apple is doomed because they aren’t doing X, where X is a move by the competition which eventually fails. Why are we even entertaining such nonsense?

 

[IG88]

I'm not actually - I have an MS in Comp Sci and studied machine learning. The 3D mapping of your face from the dot projector uses neural nets and machine learning. The reason so many images of faces were used by Apple to train the neural net that's used for processing images is to distinguish between animate/inanimate images, variations of the same face (i.e. with glasses, a hat, facial hair, etc), differentiating between a face and the background, etc. That's what training a neural net involves. You don't just throw some arbitrary number of points at a face and then determine 'yes, that's so and so' and then decide to change the parameters of the network (number of points projected would be one of those parameters).

You're still starting with a point cloud and doing math. I never said the math was trivial. I definitely respect your MS in CS. I have a BSEE and PE in Control Systems.

The fact remains that if Apple specs 30,000 dots, 29,999 isn't going to invalidate the verification. There is a tolerance on the dot projector, and it's not a hard logical leap to make that Apple perhaps relaxed the tolerance somewhat, without affecting their stated accuracy.

 

[Tech198]

Lets see: Apple was struggling with iPhone X production with the TrueDepth sensing

 

[john123]

In this particular regard, given that every single year in the period leading up to an iPhone launch there are dozens of just flat out falsehoods from media outlets (likely fed information by stock manipulators) I don't lend any credibility to reports that can't be bothered to include any technical information. The fact that most of the tech "journalists" these days do little more than demos and reviews doesn't help much.

If you see a pattern from the same people year after year, you tend to start paying attention to it. Do yourself a favor and look at the last 5 iPhone launches. Without exception every single year the same "supply chain" analysts and "sources with knowledge inside Apple"come out of the woodwork in the weeks leading up to launch to tell everyone how thing A has changed, or somehow the manufacturing process is still in flux mere weeks before millions are to be sold. It's pure ****ing nonsense to even entertain the notion that hardware for a product that has to be manufactured in the millions is somehow still in flux less than 6 months before production ramps. That's lunacy.

Interesting. We have a very different take on the rumor mill. I actually think the rumor mill overall does a fantastically good job of predicting what an upcoming device will look like. Like anything in predictions, there is signal to noise, and I think the signal to noise ratio is actually pretty good. But that's a matter of perspective I suppose.

I agree with you that there is some ass covering that goes on, and "in flux" gets used too often in lieu of the more accurate and truthful word "uncertain."

 

[kdarling]

I just ask that the rest of the competition be held to the same standard as well. Why is a product like face ID (or even Touch ID back then) treated with so much skepticism even before it is released, while other products like the google pixel praised to the high heavens, even as problems abound?

I think it's partly because no other company puts its nose in the air as much as Apple, with its pompous Ive videos, shirt tail wannabe-cool stiff execs spouting phrases like "incredible" and "magical" to the point that they're now a parody, and acting as if they invented or own every idea around.

It's pure ****ing nonsense to even entertain the notion that hardware for a product that has to be manufactured in the millions is somehow still in flux less than 6 months before production ramps. That's lunacy.

The very first iPhone had audio parts changes a month or so before sales launch due to field testing. It also got a glass screen just a week before sale. i think it's very Apple to continue to make last second changes.

Not to mention quite often having initial production difficulties. Remember when Foxconn workers almost struck because of being untrained on how to make black iPhone 5 units without cosmetic flaws? (Apple sold them anyway.)

After all, isn't being able to wake up ten thousand workers at midnite to make a "last second change", a main reason it's claimed they need Chinese workers?

 

[Ladybug]

I never made the claim that Apple was lying. Only that I expected the denial whether the story is true or not, and I'm sure Apple has understated the accuracy of FaceID from the beginning just in case problems arise or to account for variance between units, so it'd be hard to prove that any false claims have been made.

I don't think Apple was ever in danger of false claims, because even if this story was true, they had plenty of time to correct their website and make an announcement prior to preorders. I think Apple would have done that if they needed to.

 

[DeepIn2U]

I'm not actually - I have an MS in Comp Sci and studied machine learning. The 3D mapping of your face from the dot projector uses neural nets and machine learning. The reason so many images of faces were used by Apple to train the neural net that's used for processing images is to distinguish between animate/inanimate images, variations of the same face (i.e. with glasses, a hat, facial hair, etc), differentiating between a face and the background, etc. That's what training a neural net involves. You don't just throw some arbitrary number of points at a face and then determine 'yes, that's so and so' and then decide to change the parameters of the network (number of points projected would be one of those parameters).

Regardless of your MS in Comp Sci studying machine learning ... was it focused on Neural Nets or just "machine learning"? both are terms used (machine learning 5yrs+ vs neural net 2yrs) in the public eye, yet the majority of the public, although we can discern there is a difference between the two given the context their used specifically, I suspect that IG88 meant is that 4yrs from now Apple will still through this term around and it's user will glass over the ears.

not unlike:
Retina Display
(that magic number where somewhere close to 320PPI is where the human eye cannot discern clarity of some crap).

recall how long Apple was throwing that hype/jazz around for while the competition beat Retina like a rag doll over the head? Now we hear Super Retina.

I, no degree in Machine Learning, thought the whole point of calling a Neural Net processing was that machine learning was linked to a network of computers that can act as one ... kinda like SETI project with data crunching. I'm sure I'm heavily wrong on that but I do think NN is being tossed around without real credibility yet who am I to know.

 

[iamthedudeman]

Are you contesting that the Mhz myth is true? That is to say, no matter what a higher clocked cpu is going to be faster regargless of the architecture?

I don't know why I had to dumb this down, because you seem fully aware of what I was getting at.
[doublepost=1509052142][/doublepost]
They're all completely different hardware and software implementations of the same concept. But just because they more or less fulfill the same end goal, does not make them comparable. I read the verge's article, but for some reason most of the tech "journalism" world can't seem to understand that, just because to an end user something seems similar, it doesn't mean it's the same technology.

To call Kinect and FaceID essentially the same thing because they had their roots in PrimeSense is to fundamentally ignore the engineering differences (both hardware and software) between the two. It's intellectually lazy, and at worst extremely misleading.

To suggest FaceID is just a mini kinect is so far off the mark it's cringeworthy. Kinect never had to perform in the timeframe FaceID does, not to mention the integration into the Secure Enclave. In your description Apple simply miniaturized and bolted on a kinect, but that fundamentally ignores the security protocols that were developed hand in hand between software and hardware. Apple went so far as to encrypt the traffic between the individual chips in the FaceID camera array assembly so you couldn't simply sniff out the packets with a probe (provided you have the physical device). This has to work every time, with timing that has to operate under the human threshold for human perceptiveness. The Kinect never had to do anything like that, it could fail and simply restart the game. There is far and away more that went into FaceID than a simple natural progression of what the Kinect was aiming for.

Actually that's not true. Its the same. What you just wrote is as you put it 'cringeworthy'. Stop and put the koolaid down. Stop being a armchair quarter back. I actually worked with Microsoft Corp as our marketing firm have a bid for the next five years to promote their technology. It was developed with unlocking window PC's in mind before it was used for a game console. Intel also licensed the technology from Prime-sense and later Microsoft developed their own off shoot of the Kinect technology when Apple bought them.

Not only is Face ID similar, it's almost identical to the Microsoft kinect technology. What you think that the kinect technology was only used for games? lol.

Yes it does have to perform in the same time-frame and does, actually it has to perform in a faster time-frame to face ID. And in the game it has to constantly track movements by the second! Where as Face ID only has to track a persons face while you are staying still. Which is harder to do? You tell me. If fact it is more complex than Face ID, as there are some additional tricks used by the Kinect. Its lens is astigmatic – meaning it has a different focal length horizontally and vertically. This gives it two readings per pixel, as opposed to Face ID which only has one in the vertical since it is being used while stationary.

You think Apple invented the dot matrix that recognizes someones face? Hahah. Oh boy. Can you say distortion field. Yes believe it or not Face ID is based on technology from 2009. Almost exactly.

Microsoft licensed Primesense's technology back in 2007, yes 10 years ago. Secure Enclave. Yeah ok. What do you think Windows Hello is? Windows Hello uses Kinect Technology to log on to accounts and your laptop and tablet devices that use Windows. It is also used for business for transactions, and for security. It is every bit as secure as Face ID as it uses the same security algorithms, not similar exactly the same.

The only difference is Face ID uses their propriatary chip to unlock a device in real time and Windows Hello uses a chip algorithm or code specific to each device to unlock the device. It's like using the same lock with a different key.

Stop being a Apple zombie. Wow. Amazing the responses on this thread. This is why Apple does the things it does because brainwashed people like yourself make up magical theories as to why Apple does or does not do something. When Tim says it's 'magical' like a unicorn you actually believe it. Hahaha.

Oh boy.

 

[Keirasplace]

Actually that's not true. Its the same. What you just wrote is as you put it 'cringeworthy'. Stop and put the koolaid down. Stop being a armchair quarter back. I actually worked with Microsoft Corp as our marketing firm have a bid for the next five years to promote their technology. It was developed with unlocking window PC's in mind before it was used for a game console. Intel also licensed the technology from Prime-sense and later Microsoft developed their own off shoot of the Kinect technology when Apple bought them.

Not only is Face ID similar, it's almost identical to the Microsoft kinect technology. What you think that the kinect technology was only used for games? lol.

Yes it does have to perform in the same time-frame and does, actually it has to perform in a faster time-frame to face ID. And in the game it has to constantly track movements by the second! Where as Face ID only has to track a persons face while you are staying still. Which is harder to do? You tell me. If fact it is more complex than Face ID, as there are some additional tricks used by the Kinect. Its lens is astigmatic – meaning it has a different focal length horizontally and vertically. This gives it two readings per pixel, as opposed to Face ID which only has one in the vertical since it is being used while stationary.

You think Apple invented the dot matrix that recognizes someones face? Hahah. Oh boy. Can you say distortion field. Yes believe it or not Face ID is based on technology from 2009. Almost exactly.

Microsoft licensed Primesense's technology back in 2007, yes 10 years ago. Secure Enclave. Yeah ok. What do you think Windows Hello is? Windows Hello uses Kinect Technology to log on to accounts and your laptop and tablet devices that use Windows. It is also used for business for transactions, and for security. It is every bit as secure as Face ID as it uses the same security algorithms, not similar exactly the same.

The only difference is Face ID uses their propriatary chip to unlock a device in real time and Windows Hello uses a chip algorithm or code specific to each device to unlock the device. It's like using the same lock with a different key.

Stop being a Apple zombie. Wow. Amazing the responses on this thread. This is why Apple does the things it does because brainwashed people like yourself make up magical theories as to why Apple does or does not do something. When Tim says it's 'magical' like a unicorn you actually believe it. Hahaha.

Oh boy.

All I hear is blah blah blah; sarcastic and condescending blah's. But still blah's.
Lot of the things your saying are pure horse manure, go back and do your homework.

 

[iamthedudeman]

All I hear is blah blah blah; sarcastic and condescending blah's. But still blah's.
Lot of the things your saying are pure horse manure, go back and do your homework.

I am condescending? I just respond in kind. I wouldn't call 'cringe-worthy' and intectually lazzy not condescending.

I worked with microsoft in marketing the technology. Sorry to burst your distortion field. If if is horse manure please explain how. What homework do I need to do? Be specific.

My post was very specific. It's blah blah to you because you do not understand the technology. I do. Everything I said was factual and can be confirmed.

They are more alike than different. Both use the same technology to unlock your face, bio-metric data is stored in a similar fashion, locally on the device. Only difference is one uses a chip stored locally and one uses on the enterprise a server and non-enterprise a software algorithm.

Face ID should be faster since the the secure enclave is a better solution to a software algorithm. Faster using a chip on the device than a software solution. Windows Hello is clocked at two seconds or less and Face ID should be a second or less. It should work very, very well. But the technology is very similar. Microsoft Surface devices use a chip similar to Apple's secure enclave provided by intel and that is in the one second or less range.

These are the same documents we got from Apple and Microsoft when marketing their products. See for yourself.

https://images.apple.com/business/docs/FaceID_Security_Guide.pdf

https://docs.microsoft.com/en-us/wi...o-for-business/hello-biometrics-in-enterprise

https://docs.microsoft.com/en-us/wi...experiences/windows-hello-face-authentication

This.

https://www.theregister.co.uk/2017/08/17/apple_secure_enclave_decrypted/