Skip to content
2DIAMR – 2D Is Against My Religion!Subscribe

Expo & conference podcast· 18 min

Gaussian splats on web maps: why centimeter accuracy doesn't matter – Jaroslav Poláček, MapTiler

Watching here? Subscribe on YouTube.Get the next 2D Is Against My Religion episode in your YouTube feed.

Subscribe

Listen to the audio

Summary

Recorded live at INTERGEO 2026, Jaroslav Poláček of MapTiler tells Michal Gula and Tomáš Barnáš how developers put Gaussian splat models from drone imagery on top of a satellite base map for hotels and real estate, why visual georeferencing without centimeter accuracy is enough for that audience, how MapTiler streams splats with LODs, and how MapTiler compares with Google and Apple Maps for developers.

Key ideas

  1. 1According to Jaroslav Poláček, MapTiler is a Swiss company with a complete mapping stack for developers who build web and on-prem map applications; according to Jaroslav, 500 million people look at maps built on it every month.
  2. 2Developers are asked by hotels and real estate clients for something more than an image or video, according to Jaroslav, so they build photorealistic 3D maps: a Gaussian splat model from drone imagery on top of a satellite base map.
  3. 3MapTiler's 3D editor lets users georeference a splat visually, without centimeter accuracy; according to Jaroslav, anyone who needs centimeters would rather work with point clouds, while tourists and apartment hunters care about visual quality.
  4. 4Every uploaded PLY file goes through a pipeline that creates LODs, optimizes the model for the web and saves it in a streaming format; according to Jaroslav, at the lowest level only 0.05% of a splat model's data is loaded to the device.
  5. 5According to Jaroslav, for a developer the technology of MapTiler, Google and Apple Maps is pretty much the same; what customers notice is the price and a technical support team of humans who help with project-specific questions.

Quotes

“Every month there is 500 million people looking at our maps.”
Jaroslav Poláček ·
“If they need centimeter accuracy, they would more likely prefer point clouds or working with point clouds.”
Jaroslav Poláček ·
“At the lowest level of Gaussian splatting models, there is only 0.05% data loaded to the device.”
Jaroslav Poláček ·
“For developer, there's no difference.”
Jaroslav Poláček ·

Quoted in tech guides

Questions answered

Can you put Gaussian splats on a web map?
Yes, according to Jaroslav Poláček: MapTiler customers place Gaussian splat models from drone imagery on top of a satellite base map, for example for hotels and real estate websites.
How do you georeference a Gaussian splat in MapTiler?
According to Jaroslav Poláček, users search the location in MapTiler's 3D editor, adjust the scale, and the model is placed on the terrain; it is done visually, without centimeter accuracy.
How does MapTiler stream large Gaussian splat models?
According to Jaroslav Poláček, every uploaded PLY gets LODs, web optimization and a streaming format, so at the lowest level only 0.05% of the data reaches the device.
Why choose MapTiler instead of Google or Apple Maps?
According to Jaroslav Poláček, the technology is similar for developers; customers notice the price and a human technical support team.
Can other MapTiler users see the data I upload?
No, according to Jaroslav Poláček: uploaded data remains the customer's own, only they can build map applications on it, and there is no marketplace for sharing data sets.

Transcript

Full transcriptabout 12 min readShow

Transcribed automatically (ElevenLabs Scribe) and lightly edited. It may contain small errors in words or speaker names. Click a time to play the audio from there.

Michal Gula: It's the second day on the INTERGEO and, we have seen multiple people, coming to our, our, podcast studio with very different, impressions of, of the INTERGEO. What are yours? How do you like th- this years?

Jaroslav Poláček: Yeah, I, I really like this INTERGEO. I, I just realized that my fi- first INTERGEO that I attended was 14 years ago.

Michal Gula: 14?

Tomáš Barnáš: 14.

Jaroslav Poláček: Yeah.

Tomáš Barnáš: Okay.

Jaroslav Poláček: So-

Michal Gula: Are you that old or you-... di- did start so early?

Tomáš Barnáš: Right after getting 18.

Jaroslav Poláček: Yeah. So it was, yeah, first experience. Now it's, it's changing a lot, so lots of, lots of new devices, lots of, lots of hardware, new, new technologies coming. So it's... And it's exciting to see how many new things were developed in, in such a, say, short, short time.

Michal Gula: Jaro Poláček from Interge- from Interge- from, from MapTiler is, is our guest here. you also have a booth here?

Jaroslav Poláček: not yet. no, not this, not this year. we are going to, to other, say, events this year, but, uh-

Michal Gula: So y- you are still scouting?

Jaroslav Poláček: Yeah. I'm just, walking around and see what's, what's new.

Michal Gula: Scouting 14 years.

Tomáš Barnáš: 14 years, yeah.

Michal Gula: Well, what can you tell us about, MapTiler? What do you guys do?

Jaroslav Poláček: Yeah. MapTiler is, we are a Swiss company. we, we have, complete mapping stack for developers, so the developers can build map applications for, for web. and, yeah, for on, on-prem, on-prem applications. And one of the most exciting things we, we have are, support for GeoSplats. so that's why I'm, I'm here today, to see-

Michal Gula: we, we can speak about that, but maybe before that, can you tell us, like, is there any company we may know, who is using your, your platform for the maps?

Jaroslav Poláček: Yeah. There are lots of, lots of companies around the world. So, every month... So it's, it's for, it's for developers. So developers create applications, map applications with, with our product. And every month there is 500 million people looking at our maps. So-

Michal Gula: Okay

Jaroslav Poláček: it's quite lots. It's, it's hard to say one of or few companies using our, our platform, but there are, there are lots of, lots of them from around the world.

Michal Gula: I'm just guessing that you are not, like, collecting the data, yourself. You are buying the data from somebody?

Jaroslav Poláček: Marek? Yeah, we are not collecting data, so we... But, we have, a data team who, who collects, it from, from different sources and, and see what's, what's new to, to provide our customers, say, the best, best data, best, best accuracy or latest, say, satellite data.

Tomáš Barnáš: So that I wanted to touch, like, are you mainly using the satellite data or only? Because there are also, like, countries which had now, like, mm, digi- digitization of the countries done by planes, yeah, plane LiDAR and all of that. So you are pure focused on, on the satellite?

Jaroslav Poláček: We collect both, satelli- satellite data and, and also, the vector data from, say, open resources and, and some, national, national, national agencies. So we combine them together. we always make sure that it has the quality so there is no, data that, doesn't match the, the, the real, real data in, in, in the field. So, like fake buildings or, so we always, always check if, if, the quality of data is, is high.

Michal Gula: You already mentioned that, you're working also with the Gaussian splats. I don't think that your client, clients wanted it. W- was your idea to implement this?

Jaroslav Poláček: Surprisingly, yes, they are interested. So developers are those who are asked to just create 3D realistic application-

Michal Gula: Okay

Jaroslav Poláček: for, for their clients. So for example, hotels or real estate will come to them and ask them to, to create something different than, just, having an image or video on, on, on the hotel's, website. So they create photorealistic, 3D, 3D, 3D map using Gaussian splatting and some they place markers and-

Tomáš Barnáš: Also based on the satellite images?

Jaroslav Poláček: It's from, from drone imagery.

Tomáš Barnáš: Oh, drone.

Jaroslav Poláček: Yeah, yeah, yeah.

Tomáš Barnáš: Okay. Okay.

Jaroslav Poláček: So... And then we-

Tomáš Barnáš: So both then

Jaroslav Poláček: combine the satellite imagery on, on, as a base map. And, and Gaussian splatting model on top of it.

Tomáš Barnáš: Okay. That I understand. I was just like, "Oh, that's the first time I hear that someone is trying to play with the satellite images and turn them into Gaussian splats." So you have these applications.

Jaroslav Poláček: Yeah.

Michal Gula: And, s- so tell me, like, what is the... Do you have also like, like a 3D data of the buildings from, from the... Like at least the basic? Of, of course we cannot get the, all the details, but at least like the LOD1 or, or whatever. Do you have a 3D data from the buildings?

Jaroslav Poláček: Yeah, we, we have it. So it, it represents a number of floors of, of those buildings. So there are not, say, no, no details-

Tomáš Barnáš: Okay

Jaroslav Poláček: about the structure of, of the building, but, it's, it's a box. So we, we have it and it's already, it's already-

Michal Gula: Like all over the world or do you have like-

Jaroslav Poláček: Yes, globally

Michal Gula: areas where you have and where you don't have?

Jaroslav Poláček: Yeah, globally. Yeah.

Michal Gula: Oh, okay. And, are there any cities or, or sections where you actually have like photogrammetry data of, of the, of the 3D structures or?

Jaroslav Poláček: Not yet. It's, we are, searching for, for this data, but we currently don't have, the s- the structure or, or, uh-... or colors of, of, of the buildings. So it's, it's, it's always, say, gray boxes. And the photorealism we are adding is, is now, through splats.

Michal Gula: Have you ever thought about, like, allowing your clients or potentially like, like a any reality capturing company to just send you the data and to include in, in those maps? Like, like for example, we will scan the portion of the city, and we have, like license from the city to, to use it however we want. And we decide to send it to you so, so you can have it in, in your maps.

Jaroslav Poláček: Yeah.

Michal Gula: But let's turn it around. We decide to sell it to you.

Jaroslav Poláček: Sorry, sorry. Sorry. The data is still remain you, your, your data, so it's, it's not like, uh-

Tomáš Barnáš: Okay

Jaroslav Poláček: when you upload something to us, it's, it's becomes public. So it, it still remain your data and you... only you can build, map applications on top of it. But yeah, if you are nice enough you can just let us know and, yeah, we can, we can, we can add it there. But, currently we, we only support customers' data, and one- once customer upload the data, only they can, they can have access to it and they can build map applications.

Tomáš Barnáš: So you are actually keeping it always separated. Like if I as a customer-

Jaroslav Poláček: Yes

Tomáš Barnáš: start to use your, your database, I can do there whatever I want, but you don't allow another customer to touch my data until, for example, I don't... If I allow, then will you bring it to the database and everybody can do it? Or yeah?

Jaroslav Poláček: Currently there's no marketplace-

Tomáš Barnáš: Okay

Jaroslav Poláček: for, for it. So it's, it's something like this. We, we, we were asked several, several times for, for this feature. but currently it's, it's not, it's impossible to share, a data set, uh- between users.

Michal Gula: How do you solve the, the georeferencing of, of Gaussian splats?

Jaroslav Poláček: We have our own georeferencing tool. We have 3D e- editor where we, where we enable users to georeference it on, on the map. So they can simply search the location where the model is, is located, and then they will change the, change the scale and, and the model is automatically placed on, on the terrain.

Michal Gula: Do they do it just by the look at it? Like, just like really-

Jaroslav Poláček: Visually

Michal Gula: more or less visually?

Jaroslav Poláček: Visually.

Tomáš Barnáš: Okay.

Jaroslav Poláček: Yeah, yeah, yeah. So there is, there is no, say, centimeter ac- accuracy. so it's up to user to, to georeference it-

Tomáš Barnáš: We also-

Jaroslav Poláček: in, in a detail they, they need.

Tomáš Barnáš: We also discussed it, multiple times today that, like you said, centimeter accuracy. There is always importance to know like what I need, and maybe in this scale, the centimeter accuracy doesn't play a role because you are speaking about world scale, right?

Jaroslav Poláček: Exactly. Exactly. If, if they need centimeter accuracy, they would more likely prefer point clouds or working with point clouds. In this, in... or in our case, it's, it's more about, say, visual representation of reality for, for to public, or to tourists or, yeah, people who are searching for a new apartment. So they don't, they don't care much about, about the centimeter accuracy. They, they are looking for, say, visual qua- quality or visual representation of, of the place to better imagine how-

Michal Gula: Is, is this feature constantly being used by your existing users? Or have you already like had a client that came to you just because of this feature?

Jaroslav Poláček: For both. but we expect more, say, new segments coming to our platform because of, Gaussian, Gaussian splatting because it's, it brings, say, new level of, of, of realism to maps. So it's not like 2D map with, gray boxes representing buildings. It's, it's completely new use cases coming to us.

Tomáš Barnáš: So are your data, streamed on the end or do you have like, like, like LOD system or, what's the technology behind?

Jaroslav Poláček: Yeah. Every PLY file that is uploaded to us, we need to prepare for, for publishing. So we have, a pipeline which, creates LODs. it's also optimize the model for web publishing, and it save,.. and, and the final step is saving it, in our publishing, publishing formats, in streaming format. So yeah, we need to make sure that the data users upload to us, are smoothly rendered to, to users. So there is, yeah, it, it shouldn't be... The performance shouldn't be an issue. So yeah.

Tomáš Barnáš: I think performance and also the amount of data you need to use when you open it on your device, right? Like if it's for mobile devices, then you also take care of that, that not someone is opening your world map and suddenly- his data are off.

Jaroslav Poláček: Yeah, yeah. Everything's styled, everything is, has L- LODs, so we can stream only the data that user, needs to see. so for example, at the, at the lowest level, of, of, Gaussian splatting models, there is only 0.05% data loaded to, to the device.

Tomáš Barnáš: As you mentioned, like you are not, exhibiting here on INTERGEO. What would be your like ideal conference or expo, where you can find your target customers? What, what... Where it would be? What it would be?

Jaroslav Poláček: Yeah, yeah. For example, later this year we are going to Smart City Expo to Barcelona to, to show our products because, yeah, Gaussian splatting are closely related to, to Smart City use cases. And, yeah, so it's, it's one of them.

Tomáš Barnáš: So kind of Smart City target is maybe the most S- one.

Jaroslav Poláček: Yeah, smart city and also tourism. So earlier this year I was, just going to, to the biggest tourism expo, to, to see, the, the opportunity for, for, for Gaussian splatting, and it's clear that the, yeah, the opportunity is, is big in, in this segment.

Tomáš Barnáš: Mm. Do you have various, like, licensing models how you offer the data? Or, like, if I want to build, for example, like, a navigation tool for, I don't know, cyclists and then... Or, like, our area of, like, tourists, because, like, I was always thinking about it. Like, if I want to build an application where I need a l- mapping system, I found it always, like, not comfortable that I would put there, like, some maps, like, buying, cutting, putting-

Jaroslav Poláček: Mm

Tomáš Barnáš: creating the application. So this feels like a great fit.

Jaroslav Poláček: Mm.

Tomáš Barnáš: and, yeah, h-how, how that works?

Jaroslav Poláček: We try to keep it simple. so we have one free tier, one paid tier, and one custom for, say, custom, bigger, bigger, projects.

Michal Gula: Okay. how, how is it going with your competition? Like, why would one choose MapTiler instead of, I don't know, like, Google or, or Apple Maps?

Jaroslav Poláček: Mm. Yeah. It's, it's very... Yeah, the products are, are similar. it's on, on, on the users what, what they, what they decide. But the technologies are, are similar. We have, we have geocoding, we have base maps, satellite, street. I think that's... What customer will notice very quickly is, is also the price, that it will become, much more expensive than way, when they use our, our products. But everything from, from base map, from, from tiling, from geocoding, and, yeah, the, the other stuff are pretty much the same. For developer, there's no difference.

Michal Gula: how big is your team?

Jaroslav Poláček: Currently we have, 80 people, and it's, it's still, it's still growing, so.

Michal Gula: No, the question was connected actually to, to the question before, because that might also be the reason why to work with you instead of the Google or Apple Maps, because you are much more flexible, right? Compared to these behemoths of-

Jaroslav Poláček: And achievable.

Tomáš Barnáš: Yeah. It's-

Jaroslav Poláček: Yeah, and also the technical support is one of, one of the things. So if you need something, we have, we have technical support team who can just help you- very quickly, so you don't need to wait or search, long time on, on the, on, on the internet who you should contact.

Tomáš Barnáš: But Apple has Apple intelligence.

Jaroslav Poláček: It can, it can sometimes, sometimes help, but, uh-... yeah, people who, who contact our technical support typically have some, say, specific questions and, and, yeah, related to a project, and they need human to, to help them.

Michal Gula: So the service is, I- I'm just guessing the, the cross-platform, like you can use it from computers through mobile phones, tablets, like everywhere, right?

Jaroslav Poláček: Exactly. Yeah, yeah, yeah.

Michal Gula: Okay.

Jaroslav Poláček: It can be used, yeah, for, for developing any, any, mapping application on any device.

Michal Gula: Great. do you allow for measurements on your maps? Because for example, Google Maps, you cannot really measure on, on those a- Like, I have, like, no idea why.

Tomáš Barnáš: They have this Google Earth thing. Like, it's weird, no? Like, you have two products, it's like from one company which are kind of competing.

Jaroslav Poláček: Yeah, we don't, we don't have measurement, measurement tool, so we have-

Michal Gula: Then you must know the reason for that.

Jaroslav Poláček: So I write it as a, as a insight for, for our future development.

Tomáš Barnáš: But I assume that any developer who is taking your base can make the, the measuring tool on top of it, right? Because the scale should be proper.

Jaroslav Poláček: Yeah. Yeah.

Michal Gula: Or no?

Jaroslav Poláček: I don't think it's, it's currently possible. They, they can, they can use our SDK to just add any, say, any, any content to, to the map so it match the, the use case. I don't think we can measure... We, we can show, the, the terrain and, and the shape of, of the terrain. Or something for, for s- sport applications or for sport, or for, for sport apps, but I'm not sure if we have measurement. I don't think so.

Michal Gula: So, so, but the first time you, y- you said that it's for web support. Now we are speaking about applications, so, does your platform work also, or does your data work for, with the gaming engines? Like, I don't know, like Unreal or U- or Unity or?

Jaroslav Poláček: not in most cases. So it's, it's typically for more, for web development and-

Michal Gula: Okay

Jaroslav Poláček: and mobile applications. So for every, any use case where, a map, it's, if it's 2D or 3D is, is needed, so we are the right place to, to start.

Michal Gula: We do not accept 2D maps, so 2D is against my religion.

Jaroslav Poláček: Okay. 3D. Thank you.

Michal Gula: Great. Thank you. Thank you for coming, for sharing these, insights with us, and, good luck with your, company, with your, product, and enjoy the rest of the show.

Jaroslav Poláček: Okay. Thank you.

Michal Gula: Thank you very much.

Jaroslav Poláček: Thank you.

Full show notes

Jaroslav Poláček is a Product Manager at MapTiler, a Swiss company that gives developers a complete mapping stack for web and on-prem applications. Per Jaroslav, 500 million people a month look at maps built on it. His first INTERGEO was 14 years ago.

We talk about GeoSplats, MapTiler's support for Gaussian splats: drone-based splat models placed on a satellite base map for hotels, real estate, smart city and tourism projects. Jaroslav explains why 3D buildings are still gray boxes, why customer data stays private with no marketplace, how a PLY upload becomes a streamed model with LODs where only 0.05% of the data loads at the lowest level, and why georeferencing is done visually, without centimeter accuracy. Then the competition: per Jaroslav, for a developer there is no difference from Google or Apple Maps except price and human support. Is visual realism enough when nobody can measure on the map?

Recorded live at INTERGEO 2026 in the 3DISE podcast studio.

Hosts: Michal Gula and Tomáš Barnáš. Guest: Jaroslav Poláček, MapTiler.

Share:LinkedInFacebook