To ensure that Apple was applying the best image blurring technology possible, Apple conducted an entire year of test drives through the US and Europe in 2015, where it did not publish any of the imagery collected during these drives but used it instead to improve Apple’s technology such as image blurring techniques on such objects as faces and license plates.
This might explain some part of the delay between the earliest data collection and publication.
This is still strange though - it's not like images are destroyed when they are used for testing. Is there a reason for not using these in Look Around that I've missed?
This still doesn’t explain what was uncovered with the New York images. If Apple shipped them a month after collecting them (and if Apple has spend the past 5+ years driving), then why are so many areas missing Apple’s Street View?
I thought it was obvious. Author wrongly assume those Data Collection from 2015 - 2018 were used for Maps. When they were actually done for training purpose or R&D. And it is obvious their detecting Algorithm is no where close to what Google is doing.
Look Around is so much better than Google. But Apple Maps is still far behind Google Map in terms of accuracy.
I love this type of article, and the effort put into it - but the main thing I can't help but do is wonder what kind of person is driven to exhaustively research Apple Maps over a 5+ year period.
> I love this type of article, and the effort put into it - but the main thing I can't help but do is wonder what kind of person is driven to exhaustively research Apple Maps over a 5+ year period.
I'm under the vague impression that Justin O'Beirne works in this space, so this isn't just a hobby. Can anyone confirm?
He is also listed on several mapping-related patents for Apple over the past several years.[1] It seems surprisingly un-Apple-y to allow someone in that position to publish articles detailing the flaws in Apple maps, speculate about where they get their data and how the quality problems might affect unannounced future products.
> It seems surprisingly un-Apple-y to allow someone in that position to publish articles detailing the flaws in Apple maps, speculate about where they get their data and how the quality problems might affect unannounced future products.
That tells me he no longer works at Apple. If he did, why would he need to speculate about Apple's internal processes based on what they publicly release? This article (https://www.justinobeirne.com/a-summer-of-google-maps-and-ap...) from 2017 even has him noting hits from Apple's website and tracking how long it took them to correct gaps he outlined.
Those two articles I linked were dated 2016 and 2017. Looking at his website, before that time it seems like he was mostly writing about Google Maps. So maybe he wrote about Google when he was at Apple, then started writing about Apple when he left.
There’s certainly overlap between the date stamps on his articles and his Apple patents. Not sure if someone might be listed as inventor on a patent filed after they left.
There are probably a lot of teams working on different aspects of the maps, so he wouldn’t automatically be privy to everything because he’s worked on one aspect of the product.
I will also say it took way too long to get to the point. Several points that could've been made with 1-2 examples had 5-6. It got a little exhausting.
After too much scrolling just now, I remember a similar-looking article, also about Apple Maps, from a few years ago that was probably from this same blog.
His style is way too many illustrations and a few lines of text in between. Feels buzzfeedy. Sorry, just not for me.
I still went through it until it got to the SHORTCUTS section. At that point I got frustrated wondering what this pivot was all about and how much longer this article was going to go on until it got to the point (as many here have said) of what he wants to say. Sure, evidence, but the many tables and images that I had to focus on to interpret just distract from the flow of the story. Too many! There are also too many maps trying to show coverage, they don't take too much brain power to interpret, but after the first 3 I thought "yeah yeah yeah, I fucking get it".
Obviously it's a free world to disagree, but if I were his editor I'd tell him his blog posts are ineffective because it's rambling, and if he presents it more concisely, more people would appreaciate it.
Do you also think these were bad / annoying to read?
(The reason I ask is because I started a newsletter at the beginning of the year and have been trying to emulate his style. But if people genuinely hate it, then I obviously don’t want to follow it. :))
Yes they were annoying to me, and that's why I remembered them.
Upvotes might not mean much, maybe other information addicts saw a dense page and "ooh, endorphins, upvote".
After a certain point it just got exhausting to follow. Reading is simple, but looking at images so I can agree with the conlusion he made tasks the brain.
Oh, the moat article even has footnotes that scrolls to the end of the page and you then have to click back... god damn, how!!!
If he was in a conference room presenting a slideshow (each image being a slide), how long would his presentation be? At what point does he lose the majority of the audience? At least with slides people usually point out what they want you to pay attention to, and they don't do 7 slides of examples for one point (like the "look, Google Maps has outlines of buildings!" in the moat article).
I appreciate his thoroughness, it just makes it hard to read when the reader doesn't actually care about seeing every single data point. I would prefer he give an "abstract" at the top and then go into all the details below.
2017's article from the same author [1] about Goggle Maps' use of photogrammetry and other building scanning techiques was, in my opinion, one of the most interesting HN submissions ever (its comment section[2] is also worth a read).
>...seem to suggest that Apple might be having issues algorithmically identifying POI information in its imagery.
And if that’s the case, it raises interesting questions regarding Apple’s overall mapping effort...
Really? The quote sounds grandiose, but what "interesting questions" does it raise, aside from "it's not perfect" (who would have thought) and "what else might be missing?"?
Perhaps the article has been updated since you wrote this, but it no longer mentions "interesting questions" and instead includes a footnote saying if they can't annotate these images, that doesn't bode well for their rumoured AR headset.
Seems like the vehicle switch could be well explained by a big expansion in the number of vehicles (so move to something that has lower lifetime cost).
A big expansion in vehicles might be motivated by trying to have fresh imagery for most of the country.
The POI issue could just be that they just don't have a good POI database (because they all have problems).
My first thought was the view angle. Notice that cameras are placed immediately on the roof of the van, so the blind spot around the van must be wide. Subaru is a smaller car _and_ cameras are raised high, probably to get very tight blind spot around the vehicle.
Did anyone actually finish this article? It felt like it was building towards something really interesting but could have used just a bit more editorial discipline.
Sometimes, article have low information/length ratio.
This one has a ton information of information, but still fail to satisfy because it has a low point/information ratio. It just rambles on with data points and examples.
AFAICT, the points made are:
* Apple switched from vans to subarus for data gathering. (IOW, they changed their data acquisition equipment.)
* Their new equipment must be better, because Apple seem to have thrown away all data they acquired with vans.
* Yet, they still lack tons of labels and mis-label / mis-place many buildings / store etc.
* Apple seems to be processing data faster than they used to.
... and even these points are must of the shrug / why would I care category, except maybe the fact that most cities are not labeled.
Take-away: you're better off with google street view.
> This one has a ton information of information, but still fail to satisfy because it has a low point/information ratio. It just rambles on with data points and examples.
On his front page, this article appear in this section:
> Research + Updates
> Material for future essays + updates to earlier essays
It honestly reads more like notes than an article.
> they still ... mis-label / mis-place many buildings
On the contrary, that was in the traditional map.
The author’s actual point came after that: that Look Around didn’t have the same data quality issues, proving so free of such errors the author speculated whether they’re using manual labeling.
It had gone into too details about author's life and specific cities midway and I ended up with just the conclusion that there is some chaos between different parts of Apple Maps - Look Around, POI etc.
I also felt it was building towards something really interesting.
TL;DR since most people (understandably) didn't have the patience to read through the lengthy, drawn out, repetitious build-up of false-endings in the article:
It looks like Apple threw away all data collected by their first (van-based) Look Around driving effort in the US. All released imagery is from their most recent Subaru-based re-driving of the US.
To ensure that Apple was applying the best image blurring technology possible, Apple conducted an entire year of test drives through the US and Europe in 2015, where it did not publish any of the imagery collected during these drives but used it instead to improve Apple’s technology such as image blurring techniques on such objects as faces and license plates.
This might explain some part of the delay between the earliest data collection and publication.