The CEO and CTO of the Berlin-based computer vision startup reflects on the last six months, detailing how things have changed and what the future looks like for its image-assessing AI. N.B: This interview originally appeared in Automate Pro Europe magazine 2. All information was correct at the time of publication.

What’s changed since we last spoke?

It’s been a very exciting six months. One of the major pieces of news is that we closed a 5.2 million euro fundraising round led by Ventec, which is a French VC and almost all of our existing institutional investors rejoined the round. So we’re well-capitalized for the next level and it means that we’re getting into the next league and accelerating certain efforts in certain sectors very fast.

Just to give a little bit of recap of what we do first for the readers: we build computer vision solutions, which helps not only developers but also non-technical users to use computer vision solutions in their products. For example, it can be a media organization such as a press company, which has a catalogue of millions and millions of images. For people to organize that kind of content, you either have to do it manually – which is extremely expensive and time-consuming – or you can use AI.

But think about how a press company will categorize images. They want to capture some new answers to certain stories. Let’s say we want to tell a story about Coronavirus or something like that. You want the AI to detect not only the pictorial representation of the virus but also the story of how our life has been impacted by the lockdown. What are the kind of economical and social aspects to attach a meaning to? That kind of nuanced human expression or consult to visuals?

What you need to do is retrain AI incorporating that. We have a tool that lets not just the engineers, but the people who are the storytellers – content editors or content managers – within that organization, not only use AI but train AI, according to their expectations. As a tool, we’re not limited to media.

Back to your question on what we’re doing, especially now that we have a bit more capital; we’re going at certain sectors a bit more aggressively. So media’s a sector that we have a lot of experience in, we have a lot of very happy clients, so we will be doubling down on that foot. Another sector that we’re going pretty aggressively in is the space industry with satellite companies. That’s what we have been personally busy with in the last few months.

How has the COVID-19 pandemic affected you?

It’s not affected us too much in the negative. The first few months were nervous because we didn’t know what the impact would be. Since we’re a b2b company, there was some market uncertainty but now as a business community, on the whole, I think there’s a bit more clarity. We discussed in our last interview that Mobius started as a non-office organization and we were one of the earliest organizations within Berlin to take on remote working. One of the things structurally we decided is to stay as a completely remote organization.

We did an internal survey just to understand what our staff felt and in general people like working from home. People do miss a little bit of the personal side so we’ll have events but not working moments. As a work culture, we’re completely at home for the foreseeable future.

Another aspect is that we were already an international organization to start with. Berlin has a very cosmopolitan nature. It so happened that we had people from diverse cultures already part of the organization but with the COVID scenario, we started bringing in people from other parts of the world. That’s unlike a classical model where you have headquarters and you have a subsidiary. Having a fully remote-first company puts everyone on an equal footing, which also is a good cultural thing. Decision making is not geography-specific.

It seems like we’re moving towards a better way of living with the pandemic in Europe. With that in mind, will you be going to trade shows?

We will have a presence at trade shows. It’s still tricky to go to US summits because the travel restrictions still exist but European summits are opening up. I’ll be in London for the AI Summit and there’s a summit in Frankfurt a couple of my colleagues are going to as a way of starting to attend shows again. It’s still in that hybrid phase that some of the summits are still virtual, but whenever there’s an opportunity to see people in person we like to take that.

Did you partake in any of the virtual conferences? What did you think of them?

We did participate in a few of the virtual conferences and are still speaking in a few events like Data Science Salon and Visual 1st. It’s not fully a replacement for the physical summits that used to happen. There were a lot of chance meetings like you’re standing in a queue and you meet someone and have a conversation. The virtual summits are a stopgap because of the circumstances. You didn’t have to travel and these were pretty interesting events but in the more critical trade fairs, we will be there in person.

Will you be showcasing anything at these events?

We have a booth in the AI Summit and we will have booths at a few of the summits that we’re going to. Again, that’s an interesting avenue because it’s where business lead generation happens and people coming to us is always a strong starting point of a relationship. Also going and meeting people purposefully is a strong starting point of a relationship. It’s something that we missed.

It’s also a question of signalling how genuine you are in certain communications because personally, for example, I get 30 to 40 emails every day from different services. Especially after the funding round announcement, there have been a lot more emails coming in! What’s missing in virtual communication is that it’s very difficult for me to qualify what’s a genuine outreach. Ideally, you want to spend your time with people who are genuinely a good fit and not to waste too much time on people who are mostly in a mass outreach scenario. Trade fairs have always given us that opportunity. We could go and talk to prospective clients and prospective buyers could come and talk to us. That’s a useful quality, especially at the start of a pre-Covid era.

What new developments are you most excited about?

One is how we’re spending time. I’m spending a bit of time organizationally growing the company because after the fundraising round we have a bit more capital and we need to bring the right talent into the organization. We had already managed to find and bring some of the most exciting talents all around the world into the company. Still, many positions are open.

The conversation with the new sectors is also exciting. We talked a little bit about the space industry, which is super exciting for us because the amount of value that we can unlock in that sector is massive. So we were spending a lot of time building relationships with some of the critical players in this industry, getting the technology in their hands, getting feedback about that.

And of course, improving what we already have, especially with expanded and much more renewed energy and focus with more people working on that particular problem becomes very exciting for us. So these are the three steadfast areas. This is directly contributing to our commercial growth as a business. So I will say in the last two or three months after the fundraising round, we have been putting the foundations down. Some of the early results are starting to pop up.

You announced that you’re going to be moving into the US and Europe. What’s that going to look like?

One place that we have started to hire a bit more aggressively from is the UK, especially from the business development and the business teams. Our Senior Chief Commercial Officer will be relocating to the UK so we’ll have a stronger presence starting in the next month or so. The US is something that we’re still planning. There are more decisions to be made that, again, are tied to the pandemic as it’s still not fully open for Europeans to go. There are more logistics to figure out but it’s on the cards.

Is that move a personnel shift into those countries or are you hoping to do more business with the companies from those countries as well?

It’s a two-fold effort. We want more clientele and more boots on the ground in these geographies. One amazing thing about us creating a technological product is that we can sell across the globe. It’s applicable across different geographies and extremely customizable. That lends itself to being able to localize and be applicable in various jobs. Traditionally, we had 50% of our clients coming from the US and 50% from Europe so structurally, we’re an international product. We would also like to grow the team internationally, being a truly international organization.

If we circle back to your movement into the space sector, what kind of applications are we talking about?

Mobius does two things very well. One is that the solution can be easily customized to a certain end-use case. The other aspect is as a solution, it’s super lightweight. It can be put into hardware, doesn’t require too much power and doesn’t cost too much to compute. So we’re talking with a lot of people in the advanced stages with satellite operators.

At the same time, we’re also talking with people who are on other levels in this hierarchy like those in the data distribution business and people who are providing intelligence. In this sector, we have seen three major levels of applications. In level one, people want to identify and spot certain objects of interest. For example, you want to find a shipping vessel that is lost and pinpoint it from a large amount of data. Can you spot this event happening?

The second level of applications is where we’re helping our clients detect change. Sometimes city planners use satellite data coming from different captures from different durations to figure out how the city is evolving and what can be further done about it. Meteorological scientists can also use that data to see how certain waterfronts are evolving.

The third level of applications that we see involves taking that computer vision and connecting it to scientific modelling. For example, climate change scientists take the output of computer vision and start building more complex models. Right now we’re in level one and level two applications, where there’s the detecting of certain objects, segmenting out certain objects, how it’s evolving.

That’s unlocking a lot of use cases. The cherry on top is that we can do that on hardware which is not very expensive. So, having the ability to do that in real-time becomes very important for clients.

Why are space agencies or the companies that you’re working with considering your technology?

One is the level of accuracy that we can give to these companies. The second is the ease of customizability because the use case for it in one company to another varies. The third is the speed of the solution. It can be deployed in real-time. Those are the technical factors at the core of the decision-making process, which is where we’re pretty good in benchmarks beating the competition relatively significantly.

One other thing at the core of this industry is that it’s a lot of very talented people but they’re not computer vision scientists. For them, we give an alternative where they can do computer vision without having to spend too much time building that capability in-house. They can get themselves the fastest and most accurate system and accelerate their business instead of spending too much time on expanding their computer vision capabilities.

Your AI can categorize what it sees into an idea like body positivity. How is it able to do that?

We build this technology and it’s extremely easy to teach it something new. Take people’s understanding of something like body positivity in society – we did a really poor job. Even five years ago it was very common for people to body shame and it still exists. Once we have that awareness, once we progress as a society, how do we retrain that intelligence into machine learning systems? If you don’t upgrade the machine learning systems to have a more current and more advanced form of understanding these machine learning systems will be stuck in yesterday. That’s why our ability to train these models becomes paramount.

It’s an amazing thing that a lot of organizations understand the value of body positivity, for example. The last piece of the puzzle is we allow these organizations to train the machine learning algorithms with this idea. How it works is they can come and give the machine learning system a few select examples and this is, in our notion, a good way to immediately communicate what is positive. You can also give some skewed negative images that you don’t want the machine associated with body positivity. With this very select set of images, you’re able to retrain these algorithms.

I’ll go one level down and explain how we do it more on the technical side. So we work on this class of algorithms called Few-shot learning. Our Few-shot learning work is where we build a base model and the model has seen a lot of data. The base model may be able to classify or can pick out a few things but what we do is retrain and teach new things on top of it.

To retrain on top of it you don’t need thousands or millions of images, you only need a few sets of images because it already has certain knowledge that it has seen in the past. A lot of the scientific and technological progress that we’ve made in the company is to build up this base model, which is the foundation. Some of the most talented scientists in the world are working on building this representation-based model, which is very conducive to being adapted.

Are you concerned about unconscious bias getting built-in?

It’s important because almost all models will have some form of bias. It’s an area that the people who train the algorithm should be aware of and then try consciously to avoid certain things. Among other things we strive to see as much data as possible by training so that it’s a general and wide understanding of the data and of how data is distributed.

Something I will say is that the solution has to come from both ways. One is understanding some of the problems that we’re trying to solve. Then there is expressly looking for it and figuring out if there is a certain bias and carrying on mechanisms to correct it. It’s a problem that we take very seriously. We have to constantly improve on figuring out better ways of addressing these things.

Is that one of the reasons you said that you hire a lot of different people, not just geographically?

Having diversity in ideas is important within the organization. Generally, I think this helps as someone who is building any company to have a diversity of ideas and different perspectives coming in. When a diverse team works, different ideas go through different levels of refinement and that will always help. It may not be machine learning – it can be just how to put up this marketing campaign, right? If you have people coming from a certain style working on this, then you have a team with one way of thinking and you will build certain bubbles. When you have more diversity, you will have different ideas and more things to play around with it. It’s always good to have that diversity.

Have you come up against any new challenges over the past six months?

Building any company is always a work in progress. It’s always two steps forward and one step back but I think it’s been a great few months. Again, for us, the most critical thing is the velocity and the quality of our execution. A couple of interesting things we see coming technology-wise is a lot more awareness about computer vision products coming into the market. This means more organizations are starting to use computer vision-based solutions and that’s a good opportunity for us.

That means that we also have to be fast and grab that opportunity before our competitors.  Technology is improving drastically. Even in machine learning, there are things like self-supervised learning which is coming in and super exciting both for academia and the industry. We’re spending some time working in these kinds of aspects and improving and getting the learnings coming from academia and putting it into a product. Overall it’s been it’s an interesting cycle, we’re learning a lot.

During the funding announcement, you said that you’d hit a “pivotal moment where we started to realign and execute our mission of growth and strong inter-team collaborations”. What does that mean?

I initially came from a tech sector where 20 years ago I built technology for curiosity’s sake. One thing which drew me into entrepreneurship is explicit value creation, where you create something, you give it to someone, and they can generate value with what you have produced. This is the goal of the product, essentially.

For example, we build this amazing technology that can classify all these objects but take that and put it in the hands of satellite manufacturers and I’m trying to think about what value we can unlock for them. That’s 360-degree thinking where we have to think as a unit, where we think about not only the nitty-gritty of the technology but how the technology is impacting someone’s life positively and negatively. Then we find out what is the most positive thing it’s creating and build on top of it.

That requires some of the aspects that we talked about earlier – people thinking in a diverse form, not in personal terms but in terms of what their roles are there in the organization. As a business investor, you think about the commercial outcome and the product team has to think about usability outcome. The engineering team has to think about how robust the solution is and the science team has to think about how to advance technologies. Then you bring it all together. This is completely hidden from our users who should not know any of these things and only think about the value that you can create.

There are two things that we very much pride ourselves in creating in other people’s life. One is helping people make certain tasks superhuman, in the sense that you can do this task faster and on a scale that was not possible before. Like our users in some of the press agencies could not get through these millions of images they had. Suddenly, you are given this piece of technology and that dream of indexing every piece of visual content they had is possible.

Then they have more time to do creative work, create more revenue opportunities and monetize their content. That explicit value creation is something that we’re focused on as an organization. Again, it’s always very gratifying to think of it from a 360-degree angle, where you address every aspect.

If you could only choose one goal to achieve over the next six months, what would it be?

That’s not so straightforward but as the CEO of the organization, you want to build consistency in vision and consistency in the organization and let the smart people get on and do their work. That’s my number one goal!

You can find more information about Mobius Labs on its website

Stay up to date with the most recent automation, computer vision, machine vision and robotics news on Automate Pro Europe, CVPro, MVPro and RBPro.

Related articles

Delta Selected Among the Best Taiwan Global Brands for 13 Years

Delta Selected Among the Best Taiwan Global Brands for 13 Years

Delta was selected as one of the “2023 Best Taiwan Global Brands” for the 13th consecutive year. Delta’s brand was also valued at US$544 million, a noteworthy surge of 28% from 2022, establishing a new record. “The Best Taiwan Global Brands” is organized by the...

Traditional HDR: Not Ideal for Motion

Traditional HDR: Not Ideal for Motion

A camera with HDR produces images with more visible details and more useful data in both the shadows and highlights, which is essential for automotive applications where the camera needs to provide a clear, detailed view under varying lighting conditions. However, not...

Trending Articles

Join our mailing list

Subscribe to our mailing list to receive regular updates!