Hacker News new | past | comments | ask | show | jobs | submit login
Qwant Maps: open-source and privacy-preserving map (qwant.com)
210 points by arghno 57 days ago | hide | past | web | favorite | 55 comments



> All components developed by Qwant Maps are open source so that users can contribute directly to improve their experience with the map: they can contribute either to the Qwant Maps code or to the enrichment of the data directly in OpenStreetMap.

This is an important takeaway.

One of the pain points in the OpenStreetMap experience has been lack of performance of the reverse geocoding engine (input lat/lon, get POIs/adress). Or so I hear/see. It looks like they worked around this by only letting the user click on known POIs.

One small nitpick: I find the simplified polygons to be... too polygonal (forests, bodies of water, etc) at lower zoom levels. I don't know much about this field, but that's probably a result of not picking the right points when reducing the shape complexity? A bit like what bilinear/other filtering does for raster.


The simplified polygons are a result of using vector tiles, more specifically from OpenMapTiles project, as noted in copyright notice. They need to be simplified in order to achieve reasonable size. Although we can argue how much is too much.


This is true, but you can be smarter about which points you prune from the polygon. If you take a random polygon and delete random points, you will likely end up with a pointy shape (more so if you only leave a triangle).

I don't see why vector tiles should lower the end result's quality. In the end this is strictly an aliasing issue.

Take those two zoom levels as an example. In the middle of the screen, a big portion of the forest is discarded:

https://www.qwant.com/maps/#map=7.92/50.4850901/16.1876218

https://www.qwant.com/maps/#map=8.10/50.4787168/16.2208950

This isn't a very good example because those polygons were likely distinct, but a better job could be done by joining them. I wonder what kind of data is available to the vector engine. Would the pruning/simplification be done server-side, or client-side (with the client only requesting parts of the dataset)? In the later case, I feel like a smarter ordering of the primitives, and maybe some hinting metadata, could make a visible difference already.

On the other hand, it seems that they went with splines instead of polygons for roads, which produces a smoothed-out result. I think they should probably do the same for water bodies and green areas at lower zoom levels.


The problem with OpenstreetMap based map is that there are no good reversed geocoding to search locations. For example, Osmand app is perfectly fine to use except I can't search location by postal code. Pasting the address does not work and searching only the street name leading to multiple city with same street name.

What I am doing now is searching the location in Google, getting the coordinate and searching the coordinate in OSM for offline map. A pretty dumb workaround but nothing can beat Google's resources on refining their map.

If this project aimed to be the consumer version of openstreetmap, then I would claim that reversed geocoding is the first thing that need to be fixed.


Is that not what this very article describes?

> Mimirsbrunn (or simply Mimir) is the search engine of Qwant Maps. It allows to search for the following "punctual" geospatial objects: addresses, administrative areas and points of interest. It is a web service of geocoding (or more precisely geoparsing) that matches the user unstructured text query with a specific point (or area) on the map (i.e. a latitude and a longitude). The complexity of the problem is to disambiguate user requests. The roots of this ambiguity are numerous. First of all there are a large number of possible interpretations in the world for a given mention of a place: the difficulty is to choose the most "relevant" interpretation. Then there is the natural ambiguity due to the natural language: the user often looks for a place whose the exact spelling is not necessarily known.


How does Mimirsbrunn differ from Pelias? https://pelias.io


Both are geocoders using OpenStreetMap data and backed with Elasticsearch. The main difference is how they handle geographical context, eg to determine in which city/region each place is located. Which is useful both for indexing the data and producing human-readable labels.

Pelias is using its own gazeteer (https://whosonfirst.org/) to index locations and remarkable zones. Whereas Mimirsbrunn relies only on OSM data for this task using Cosmogony (https://github.com/osm-without-borders/cosmogony), a tool that extracts official administrative boundaries and other shapes from OSM data. But the world is a complicated place, and OSM tagging scheme doesn't always lend itself to geocoding purposes. So it comes with a lot a caveats which are still being working on :)


Thanks! I can see how using Cosmogony can make sense if Who's on First is too complex but I'm afraid OSM doesn't have enough coverage of administrative boundaries. The boundaries are mostly not visible in the physical world and thus cannot be mapped in the traditional sense.

I wonder if there's an open catalog of boundary data analogous to https://openaddresses.io/


Search by postcode (or address) would be geocoding not reverse geocoding.

Reverse geocoding takes a location and returns an address.


> The problem with OpenstreetMap based map is that there are no good reversed geocoding to search locations.

Yes, this problem does exist. Ultimately, probably Google is the best entity in the world that can map search queries to the stuff you want, but qwant maps adopting mimirsbrunn seems like an improvement from my testing.


And this problem will continue to exist if people continue to use Google Maps. You need a large amount of data (the map search queries of the user and geolocation data as you travel while using the map app) to refine and improve this. But new map apps will initially offer a poor user experience as they can only improve once people start using it. But people stop using these apps because of the poor user experience. And thus the gap between Google Maps and others keep growing as people continue to use Google Maps, and keep giving it more data and ignore the other map apps.

That is why Apple has now begun to collect geolocation data from even other apps ( https://thenextweb.com/apple/2019/06/10/ios-13-will-show-you... ) in the guise of "transparency" - the data is really to improve their own map apps


> the data is really to improve their own map apps

Baselessly claiming Apple silently uploads locally collected data is quite a leap of imagination on your end, do you have a source for the uploading part?


It's not that big of a leap of imagination when you realise that the data is shown to the user on an Apple Map and that Apple has been collecting location data since iPhone was launched, in many different ways.

- Apple is collecting location data, but not YOUR location data ( https://www.techradar.com/news/phone-and-communications/mobi... ).

- Apple tweaks privacy policy to juice location tracking ( https://www.imore.com/apple-posts-qa-location-data )

(Note the dates on these articles).

Moreover, all this "feature" of collecting data on app usage through "screen time" or the feature of getting location data from other apps to "inform" the user is just a very reasonable excuse by them to collect more data and profile you, in the guise of "transparency".

Having trusted Google once with a similar "Do No Evil" propaganda of how trustworthy they are, and ultimately been betrayed by them, I can recognize that Apple is following the same pattern as Google - collecting more and more data followed with a PR campaigns about how they "care" about a privacy. But despite all this PR about "transparency" note that there is no way to opt-out of any such data collection and sharing!

(No, I am not an Apple basher - I have and use their products and actually advocated for it a lot with others. Still do under some circumstances. But I am losing faith in them ... ).


Link to the live map: https://www.qwant.com/maps/


Thanks! It's absolutely baffling that they would write a blog post introducing their map and not include a link.


I've been looking for it too, with no luck! Thanks for the link


I hate to be "that guy", but in their enthusiasm I think the writers forgot to account for the "what is this?" demographic. I know what OSM is. I understand that Google Maps tracks me to no end. I'm curious what exactly this is in a nutshell, and it would be good for one of the early paragraphs. It looks like a UI on top of OSM, but this wouldn't be the first. I assume there's something else special here.

I appreciate any work in this area. I like the idea of a self-hosted map, if indeed this is open source to that extent.


This is cool, but how is it privacy-preserving? The server gets to see my exact search terms, and also which tile areas I'm zooming in on.

Is it just that it's not linked to an account? If so, openrouteservice fits the same bill: https://maps.openrouteservice.org/


Trying out the live map, it does seem like "just" a much, much nicer interface to openstreetmap. And it does seem to do routing, but openrouteservice is something that I use when I plan a multi-day hike in the mountains, while qwant knows/shows the opening hours of shops around me, more like google maps.

The two seem to be quite complementary. Each seems to beat google maps in their own arena, at least where I live.


StreetComplete has been collecting business hours for OSM, that is likely where this data is coming from.


We're getting to the point where mapping is becoming a simple apt-get install away, and that's a great thing.

Projects like this are important not so much because of the service they provide, but for the software that's being open sourced.

For example I was able to set up a mapping server by compiling (only to get the latest versions) mapnik and osm2pgsql, downloading data from OSM, and voilà... Next step is way finding, and there are a few options available.

Now I have another one to look into ;-)


I really tried to use qwant, but their exact match operator (« ») is unusable from a spanish keyboard, and guess the same happens for other languages, while "keyword" seems more reasonable. I've tried to contact them many times through many channels, never had a reply.

Qwant is not bad as a google alternative, but if this simple things are not polished and it's basically impossible to give them feedback, I'm not even gonna try.


I used Qwant for a while (lite, the main version is so cluttered), but found the results to be hardly usable.

I do hope they manage to stay afloat though, as I am happy about any Google challenger.


Results in spanish are not bad. For the time I used it it was useful. But there's the problem I mentioned. I use operators a lot and I can't live without them. I don't want to use alt+numbers just for the basic one, and also, everything seems or not translated to spanish, or only half-translated. I don't really get it, since you have Spain + most latin america as a market.


Same here. But now I've been using Duck Duck Go for a while and they are good enough to be my main search.


Found it on a US Mac keyboard as option-\ and option-|, so not that bad to type, just never have typed them before.


I could recommend mapy.cz (there is English language interface).

https://en.mapy.cz/

They are based on topographic map with very high accuracy - better for biking and walking in the forest. They search the route for road bike and mtb.

There is also an Android app.


Glad it has bicycle directions, since that's my main use personally.

I just tried a search, and it got a POI location (a Minneapolis park building) right even though I misspelled it, which is something other non-Google maps are weak on in my experience. But weirdly I couldn't get it find my home address (also Minneapolis), either with street number/name/direction, adding the zip code or in full.


OSM-based geocoding often fails if you include the house number (because that number almost certainly isn't mapped in OSM). Try removing that.

obPlug: if you're after OSM-powered bike directions, I run https://cycle.travel/map


That makes sense as an explanation. But it means it's hard to use in a city that has long streets with the same name the whole way.


This may work in France, but it’s shockingly bad in New York.

I’m unable to find addresses that it doesn’t think I mean Brooklyn or Queens instead of Manhattan. If I replace New York, NY with Manhattan, NY in the search, to try to stay out of Brooklyn or Queens, it ends up showing addresses in Europe instead.

For instance:

    144 W 48th St, Manhattan, NY
Goes instead to:

    144 Rue Frédéric Manhès (Champ-sur-Drac)
I haven’t deciphered how the search works to take me even to famous addresses in New York City, unless the address is in the place’s name.

If I try to find a hotel called “Grande Cloche” in Brussels, it’s the top result:

    A la Grand Cloche
If I type Hilton NYC, doesn’t work, but Hilton New York does. Seems like it’s ok with literal names, not ok with addresses or geo names.

Of maps that work well for location finding, directions, and transit, in both Europe and America, Apple Maps seems relatively privacy preserving.


The accuracy is terrible, try searching for "Bitzighoferstr 6 6060 Switzerland" and you get "Embassy of Switzerland in Moscow". https://www.qwant.com/maps/place/osm:node:698892171@Embassy_...

It should return: https://geocode.xyz/Bitzighoferstr%206+6060+Switzerland "6 Bitzighoferstrasse, Sarnen, Switzerland 6060 / 46.90353,8.2472 "

Even adding the latitude,longitude manually for reverse geocoding returns some strange results: https://www.qwant.com/maps/place/osm:node:1024510041@353_km#...


Trying to get from east London to Euston on foot, it suggests I walk on the Thames for about 60% of the journey.

I'd love to be able to walk on water, but alas, I cannot.

Guess it's still very much Beta.


Isn't it routing you via a Ferry?


It seems to be [1].

Perhaps there's a miscalculation for when to use ferries.

If I wanted to get from Canary Wharf to Euston without a car quickly, I would take the metro. If I want to walk, I don't want to use a ferry.

But of course, sometimes it's essential to use a ferry to get somewhere, like an island. Perhaps that's why walking directions include the possibility to use a ferry.

[1] https://www.qwant.com/maps/routes/?origin=street:osm:way:193...


The metro and other public transit is difficult to include in the directions because it's highly time-dependent given operating hours and transfers (for that, there's OpenTripPlanner). In addition, it's difficult to manage such complex data correctly in OpenStreetMap (as opposed to separate GTFS files and tools). OsmAnd for one is trying to get us there with a beta release of public transit directions based on OSM data: https://osmand.net/blog/osmand-3-3-released


It's certainly using the ferry route, but it's configured for on foot since it doesn't appear to support transit so I assumed it's an error and it thinks the ferry route is something I can walk?

Anyway, my main takeaway is that they've made the OSM maps look really nice, I struggled with OSMAnd to replace GMaps.


So this is basically a UI layer on top of OpenStreetMap?

Would be cool to see Android/iOS versions -- I rarely look at a map unless I'm navigating somewhere, usually in the car.


I think "basically" may be an understatement. With OSM you get "only" (or "as much as") data. The rest of the stack is up to you - whether you use some ready-made open source solution or roll your own.


Have you heard of Osmand? It sounds similar to this project


Maps.me. You can get it for Android on Google Play or on F-droid (listed as "Maps"). They also have and iOS version.


On my aged Nexus 7 (Android), running in a Firefox tab, it's useable. A bit slow, maybe due to bottlenecks in the pipe down to T-mobile here in California.

https://www.qwant.com/maps/


It's quite nice, although I prefer basemap style of maps.sygic.com - a similar venture. It's much more legible.


Hosted on Azure. Pushes an Android-browser user hard to install the app.

Based in Czechia: https://travel.sygic.com/en


Hey! It's actually good! I'd use it :)


Quant Maps search seems to be quite lacking. Searched for a large store in my city, where I recently drove using Google Maps, and it can't find it. It just responds to a match to the city name. When I used just the name without the city, it found a pub half way around the world with exact name match.


Looks like this crowd would love BSV Maps. It puts crowdsourced map data onto the Blockchain:

https://twitter.com/justicemate/status/1144481701290078209


Why exactly is a Blockchain needed for that?


Duh... To enable the DeepSearch AI and containerization of each user account.


1) It's an experiment. Perhaps it will be abandoned.

2) Curation/Upvotes via transaction needs to be on chain.


2) Why do you need the overhead of a blockchain to increment a counter?


It's not a counter, people pay bitcoin to upvote.


I think what folks need is download + local search.

- download everything about a large area: city or county or zipcode or similar

- do any kind of searching on the local machine within that data

This would be basically localized offline maps and would preserve privacy.


Osmand can do a lot there.


These OSM-based maps are pretty! Is there a way to set the distance and scaling in Miles instead of Kilometers for all of us USA folk?




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: