Mapillary joins Facebook

We’re here! A bit overwhelmed and under-resourced at the moment, but we’re here. As you can imagine, there is a bit of a transitioning phase for a few months as we integrate within Facebook. A big part of this has been improving privacy even further.

There are concerns that Facebook is interested in identifying faces. This is simply not the case. We’re blurring faces and license plates as soon as images hit our servers and we’re looking at other things we can do to assure privacy is protected. The features of interest are map related such as crosswalks, curbs, sidewalks, bicycle parking etc.

There has also been an extensive review of all the third party services we use to remove any issues that might result around privacy. While Mapillary has always been committed to privacy, we now have the resources and expertise of Facebook to accelerate those efforts.

1 Like

This is actually far from the reality. Data for Good is something Facebook collaborates closely with Humanitarian OpenStreetMap and OpenStreetMap more broadly on. See the current Microgrants Program with HOT for example.

The idea is not to replace OSM in anyway, but to overlay OSM data with other data layers that might be useful for humanitarian purposes. My own country was devastated by bushfires earlier in the year, yay 2020 :roll_eyes:. Data for Good maps were helpful for authorities to see the general areas where people were located and plan evacuations accordingly.

Population density maps and electrical grid distribution maps are other example where OSM data can be overlaid with additional datasets to assist decision making. HOT is regularly working on such collaborations.

1 Like

You forgot to mention that Mapillary keeps the unblurred version of all photos also.


Hi Eduardo,

Thank you for responding. Privacy is an issue, and for me an extremely important one.

This is good to read, but… Facebook has a very poor track record when it comes to Privacy… I’m working on a project where privacy is an issue… and above all, it’s a project I’m building and I want to build a system that will have the highest possible standing in that regard. And the thing is that the moment I would mention “it uses a part owned by Facebook” I’dd get into a position where I would need to defend myself… For a new service that is like starting the soccer match where the competition starts with two goals to zero for me…

So my standing in this subject might be a bit on the extreme, but I am quite certain there are more contributors out there that have this unsettling feeling since FaceBook acquired Mapillary … is Mapillary going to be just an other tool to gobble up more (meta)data… to get maximum results for FB-advertising?

I’ve learnt not to speak for others, so I’ll stick with “me” henceforth :wink:
Don’t get me wrong, what you have built at Mapillary is absolutely great work… it’s just, how can you convince me that FaceBook can not be able to access faces & licence plates (ever)? How can you convince me that FaceBook wil not (start to) gobble up (meta)data from Mapillary to build better profiles of their FaceBook/Instagram users?

And is it now so that with publishing imagery on Mapillary I make imagery “public domain”? All and everyone will be allowed to use data? So that means Google can gobble up all the 360 imagery they want to improve their coverage? Same goes for all and any commercial company or tiranic government that sees something valuable in there. There is no “brake” in there anywhere?

These are all points that worry me greatly…

1 Like

I probably can’t convince you of any of that. Facebook has received a lot of scrutiny over privacy in recent years, and rightly so. A positive consequence of this is that there is now extensive focus on privacy throughout the organisation. There are extensive controls on where and how data is stored, and who has access to it.

Mapillary is a tool for making better maps. It’s as simple as that. If Facebook was interested in personal data, it would not make sense to acquire Mapillary.

Images on Mapillary are still available under the CC BY-SA 4.0 license, same as they were before. Commercial licenses are being provided for local council, departments of transport, and GIS professionals who have a use for the map data. Again this is the same as before, but commercial licenses are now free.

1 Like

One idea for improvement: it is technically possible to edit JPEG files losslessly to some extent: specifically, single 8x8 blocks can be changed at will without quality affecting recompression. So one could just blur those blocks in the original images and save the blurred version only. Bonus: create a small file with the changes as an “unblur-diff”, and store that somewhere much safer, e.g. give that to the user.

Thank you @eneerhut for responding. Your loyalty to your employer is remarkable because actually it should be your boss (Facebook CEO) who should be giving statements on this subject.

I probably can’t convince you of any of that. Facebook has received a lot of scrutiny over privacy in recent years, and rightly so. A positive consequence of this is that there is now extensive focus on privacy throughout the organisation. There are extensive controls on where and how data is stored, and who has access to it.

It’s great to hear and read that there has finally been some awakening at Facebook about people’s privacy and how important it actually is. However, any changes Facebook might have introduced or might introduce are hardly going to convince anybody. You see, the problem is that for years and years Facebook has developed a business model based on collecting data about users and non-users (so basically everybody) in order to sell it to the highest bidder. This has created and nurtured a certain culture, a mindset, and a way of doing business in the company. It is not going to vanish over night. This is one problem. The second problem is that even though Facebook may have learned its lessons in privacy (though I doubt it, later more on that), they have only done so because they were forced to. And now, all I hear from Facebook is “trust us, trust us, trust us”. It is the same thing they have been saying as before “trust us, trust us, trust us”. So, why should anyone believe them now? They still have the same business model going on as before: the user is the product, not the client.

To me, a strong indication of why Facebook still has not learned anything, but more importantly has not changed its mindset, is the way they have handled the Mapillary acquisition. Apparently, they have no idea about how bad their reputation is or they really just do not care. They also do not have any sense of politics and good public relations. If they would have been aware of that they would have taken a different course of action to associate themselves with Mapillary. They could have become a major client (no need to disclose this fact), or they could have transformed Mapillary into a foundation which would have kept them in a safe distance to the data, or they could have just licensed the technology which would have enabled them to build their own mapping service. There were so many avenues to go down but no, it had to be a head on attack. So, instead of keeping low, which would have been the wise way to go, they stomped all over Mapillary and its contributors, flashed the cash and said “deal with it”. This behavior clearly shows their disregard for anything and anybody.

No matter how often they might reiterate “trust us”, no sane person is going to believe them.


I was still brewing on an answer :wink: A bit sharpener formulated then I would have, but the response of @GITNE says it all.

After re reading my post, I agree, that is the only honest answer you can give, I sincerely thank you for giving it.

If you need a president of the foundation, I am your man $$$

“Deal with it” is certainly not our attitude and I apologise if it might have felt that way.

Our intention at the moment is walk the talk and demonstrate through actions our commitments to both OpenStreetMap and privacy. As you mentioned, we can’t convince you, but hopefully over time you’ll see from our actions that we are deeply committed to OpenStreetMap and privacy. These are things we care about at both a personal and professional level and the best thing we can do is to demonstrate it in the coming months and years as Mapillary continues its journey within Facebook.


“Deal with it” is certainly not our attitude and I apologise if it might have felt that way.

Well, it certainly feels and looks this way. The thing with Mapillary is that Mapillary is not a company or start‑up like any other. It is different because it relies on the commitment and contributions of others. Sure, formally any acquisition is just a transfer of property rights. However, Mapillary comes with strings attached. Mapillary has a silent (or maybe rather voice deprived?) third party involved, which formally does not need to be heard but may not be wise to ignore. So, as I said, politics and diplomacy may be useful here.

Our intention at the moment is walk the talk and demonstrate through actions our commitments to both OpenStreetMap and privacy.

I totally agree with you on that, @eneerhut. It is too late for Facebook for just words to have any impact. Only actions matter now. And, I do believe you that the Facebook leadership is truly trying to mitigate recent data privacy scandals and is also trying to implement new procedures and data privacy standards in the company. However, for us outsiders it is very difficult to understand and verify what is actually going on inside Facebook and/or how data privacy standards are implemented. For now, we just have your word that privacy is a serious issue in the company and that actions have been taken.

Of course, I cannot speak for others but I think there are a handful of things Facebook and Mapillary could do to help restore trust in the community.

  1. Facebook CEO and head of Mapillary @jesolem should make a joint statement on data privacy regarding imagery collected on Mapillary and what is the way forward.

  2. Make a concrete statement on what happens or has happened to the original non‑blurred imagery. Details do matter.

  3. Externalize the blurring feature into a standalone application, so that contributors can automate this process locally and integrate into their workflow. I am aware that any weights data required for this to work may constitute a business secret. If that’s the case then I have no easy solution for this.

1 Like

Just FYI, the vast majority of president and other board of director jobs in non‑profit organizations are a money losing business for the people involved. Sometimes you are compensated for your expenses (which you have to prove meticulously) but that’s about it. In any event, any such job is going to absorb a lot of your time. Sorry, for having to disappoint you :wink:

:thinking: I think FIFA may be an exception to this rule? :smiley:

1 Like

I was also thinking in these lines @GITNE. Your point two more or less can be answered by point, but that could “ease some nerves”. And last but certainly not least: especially point three, @eneerhut in a sense already replied on that subject. When they could create a (separate) github of their (in my opinion very good) face & license plate detection I think that would give the option to those who want that option before uploading.
For those that don’t think it’s a problem, well they don’t have to use it then :wink: (in reply of the remark of @filipc on that subject)

PS: I would appreciate a json/xml log of each image with the detect data (and if possible with an id/info what the detector saw in that spot & a reliability value, added by a “dry run” option?)

1 Like

I redraw my candidature.
I was a pigeon, now I am a fierce enemy of FB as they want to destroy Flemish culture.
I wish Mapillary and FB many happy years deleting pictures.

1 Like

@eneerhut I do not envy you.

Exhibit n + 1 on why when you’re acquired you’re not in a position to promise anything to anyone despite any assurances from the acquirer. Or, from the consumer’s side, why those promises should carry no weight.

Faceboook account now required to login to Oculus devices"

1 Like

I think this is good news. @jesolem Thank you for taking this issue seriously and acting on it. While this solution may not be perfect it is definitely a step in the right direction and something to build on. And, I honestly hope that this can restore some of the trust which was lost to some contributors because there is no other imagery provider with a comparable level of quality of service to Mapillary. Mapillary has become an extremely useful tool to the OpenStreetMap project and also to many other companies and public service authorities. It would be a shame to loose it, just because big money rules.

Now to the solution itself: Since the blurring algorithm is not perfect, have you considered a shorter grace period, like 3 or 7 days, for storing raw imagery in order to still be able to validate blurs and collect training data? Because the thing is that contributors and consumers are as interested in protecting privacy as in unobstructed access to other data. One frequent example for obstructed data are house numbers, which currently often enough get falsely blurred. It would be nice if contributors could continue to help improve false positive rates.

It might also be worth while looking into minimal JPEG manipulation since there is surely value in storing any privacy insensitive portion of the imagery unaltered. Well, if not for now, then surely in the future.

Imho, a standalone blurring application would be even better because we could collect training data and blur images at the same time without the need for a grace period.

1 Like

Thanks @GITNE . We did consider shorter grace periods but eventually decided against it. The contributors who would be interested in going through their uploads to look for falsely blurred areas are also often the ones that upload a lot of images which makes it hard to do in a short time window. In the end we decided to put privacy first. To answer your earlier question about what happens to the original imagery, I hope that is clear now. No unblurred copies remain.

A standalone blurring application that contributors can run on their imagery pre-upload is a great idea. We are looking at it and will share any info when we can. cc @eesger

Thanks for your input!


Thank you for the clarification and welcome back to the forums :slight_smile:

Have you considered the idea, floated here, to blur in a reversible way - with a secret that would be encrypted with something the original contributor has (their password, perhaps)?
I often see blurs on addresses (housenumbers, flat numbers etc), opening hours and other useful data.
Short grace periods would not help in most cases.

Thanks for this clarification, this is good news!


The iconoclast mob has won.