Picterra is a Swiss startup which uses artificial intelligence to analyze Earth imagery from drones and satellites; their software detects objects (buildings, vehicles, trees, cattle, …) for fast and accurate observation and counting. We met cofounder Frank de Morsier in New York the day before the WeRobotics Global 2019 meeting. Our conversation has been edited for clarity.
Q: I’m in New York with Frank de Morsier, the Chief Technology Officer of Picterra, from Switzerland. Frank, thanks for sitting down with us. Now, maybe you could start by telling me a little bit about Picterra. What gave you the idea to co-found Picterra, how did that come about?
Frank de Morsier: Everything started in 2016, basically the cofounders, we met in our previous job, we were generally in the field of Earth observation and aerospace engineering. What was nice there was that we had already this complementarity. Pierrick Poulenas, who is the CEO, was in charge of the business development. And so we were used to working together and having this complementarity, it was not just two engineers who said, let’s go. We also had quite a view on what was happening in the US observation market. And we decided to go ahead and start a business because we wanted to reach a wider set of people. We saw there was really something to be done using machine learning, using a lot of things that were out there on the machine learning side but which were not yet accessible to many people.
So we started up in 2016 and we said OK, well let’s do it, go straight to clients, let’s start to build small things, small services, testing the market a bit. We did this for a year, a year and a half. We saw that everyone had quite different needs; we saw that with the same tools, we could serve those. Then we said, OK, if we really want to have a broad impact and offer our technology, a tool, to a broad range of people, we will need financing, so we started to look for that. At that stage we were three cofounders, and then one left. Basically, he was probably not feeling at ease with this startup mode; it was taking too long, and of course there was definitely some risk going with this. We found our first investor at the end of 2017, and with this we could really start building the team and developing the web platform we have now. So it went quite quickly. Going into 2018, we moved from basically being two to five [staff] and we are six now. So yes, we really pushed the development of this platform over the last year.
Q: Now, as I understand it, what you’re doing is applying machine learning to geospatial information, to Earth observation data. Did you start with satellite data and start adding drone data, or did you always think that you would have multiple sources of data?
Frank de Morsier: Actually, when we started in 2016, we had different services and they were already using both sources. So, for instance, we were looking at forestry, tree logging, from satellite imagery. And then, on the other side, with drones, we were looking quite closely at vegetation management along power lines. We definitely saw that the same tools can handle these different kinds of [imagery] and so we always kept the two [sources] and of course we wanted to put these both together. We did a bit of development there, how to transfer different resolutions. And that’s something we are quite aware of and we have tried to push this a bit. It takes time, for everything to be nicely integrated. But definitely, this idea of ranging from drone to satellite, it has always been there.
Q: Now maybe we could talk for a moment about your business model. It seems to me that you’ve been offering enterprise solutions for some time, which are probably adjusted for each major client. But I see that you also have a subscription model that will be coming soon?
Frank de Morsier: Yes, definitely. So it’s also something as we started these services and enterprise solutions. In 2016, once we started to really go for a tool and go for that vision where we want people to directly access these machine learning tools, we saw directly that the business model will be subscription-based, and definitely not a project enterprise solution. But in the end it’s never black and white and you have to transition from this. And so that’s exactly what we’ve experienced now. And so for instance, last autumn we were still doing enterprise things, but directly related to the platform. So we were customizing certain detectors for people and then putting the detectors on the platform and they could run it there, because all the features were not there yet. And so we will start to transition. For some projects we were basically doing some tests with some clients and then they could get an improved access to the platform. So there’s real distillation. And now we have published the different subscription levels and, in April, people will be able to subscribe to this.
Q: I also saw something about a reseller program. What can you tell me about that?
Frank de Morsier: Yes, that’s definitely more on the drone side. There are quite a lot of people there that are already reselling drones and photogrammetry software. So that amount of software for photogrammetry for instance, there we felt it was quite logical to address also to these people and that they could resell for us, basically they get a discount for certain packs of licenses.
Q: Now, let’s talk for a moment about AI [artificial intelligence] detection because I was looking at the blog post that went up today. And I found it astonishing because it is a very modern way to count how many refugees are in a refugee camp. I also saw on your site that you now have an interface for people to work on their own object detectors. I saw a phrase: “If you can see it, AI can learn to find it.” What do you mean by that?
Frank de Morsier: I think there are two major things around that. The first one is, I think there has been, at some point, too many high expectations around AI and somehow we are kind of lowering them. So I think the sentence you just said, that “If you can see it, AI can learn to find it”, is about this: it’s not about magical things, it’s about doing things over large scales, and automatically. And so it’s quite natural then that what you see very nicely with your eyes, you can start to teach the machine to find it. And often, people when they think AI, they think it will be a combination, a very large amount of data from different sources, etc. So this is all happening. But in our case it is more really about scaling up what you were doing today manually or just with your eyes.
What we realized when we were in discussions with different people over the past years is that each time, people were looking for things that were slightly different. There was someone who was looking for finding fallen trees, of a certain species, and in a certain type of background; next to rivers, others in the middle of the forest. So it was all about trees, but each time a bit of a different configuration. And then we felt we have to give the possibility directly to these people to teach the machine because today they are basically looking at [imagery], counting already. And then in this process the machine could start to learn. We decided there really is a need to make customized detections for each client.
Q: Now, you’re in town for the WeRobotics Global 2019 meeting and I saw that it’s actually a fairly small meeting of people who are very active on the ground. Can you speak about that a bit?
Frank de Morsier: Yes. So maybe first about WeRobotics. They are a nonprofit organization that really started from the fact that, there is of course an AI race, there is a drone [technology] race everywhere. What is missing [in developing countries] is local expertise in order to handle this. And so they see themselves basically as a network where they are providing this expertise. You could see it as an education program, a teaching program. And so they facilitate local expertise on drones and AI, they call them Flying Labs. So they have kind of a pool of competencies around the world, which allows these people to be autonomous and use drones and AI for good. After disasters, for some food security aspects, etc. At their global meeting tomorrow, the purpose is to put together different people from the different Flying Labs and then donors on the other side that are funding these Flying Labs and their different projects, and next to these, partners such as Picterra. Basically we provide an access to the platform to these Flying Labs and we act as a bridge between AI machine learning and these people that is really needed, where you don’t want to bring them all into becoming a data scientist, being a machine learning expert.
Q: One more question. Your clients have mostly been in Europe I think up to now. Do you see that developing? Do you see new markets opening up in the US or elsewhere? What’s your sense of that?
Frank de Morsier: Yes. It’s quite normal. So at the beginning, you work your local network, let’s say. And so of course it started there, Switzerland and Europe. And now with the platform, and also with this subscription mode, there are no real limits in terms of what we can reach there. We had quite a nice exchange with Australia, where they have strong needs about cattle, or post-disaster, such as floods. They are often quite affected by these. Africa, also different organizations active there. So no real limits. Our platform makes it easy to connect to these people. Also, the different social media channels have really allowed us to connect with people.
Q: All right, Frank, thank you so much for meeting with us today.
Frank de Morsier: Thank you very much, Sean, thanks for everything.