Computer vision application examples:
3 startup founders describe their experience

Irina Kolesnikova
May 5th, 2022

To provide specific examples and illuminate possible pitfalls, we talked to three business leaders who used computer vision technologies as a part of their products.

Business leaders, considering different computer vision applications, often have uncertainty about the struggles and challenges they could meet on their path. Unfortunately, most articles in this field generalize experiences; thus, it is impossible to see actionable details.

To provide specific examples and illuminate possible pitfalls, we talked to three business leaders who used computer vision technologies as a part of their products. Check out their own stories about the computer vision implementation main challenges they overcame and the biggest benefit they realized from their successful application.

We interviewed:

object detection

What is computer vision?

Computer vision is a field of artificial intelligence (AI) technology that helps to derive meaningful information from visual inputs like images and videos at a scale and speed impossible for humans.

The range of popular computer vision applications is very broad, benefiting multiple industries: from object tracking and autonomous systems for self-driving cars and warehouse automation to facial recognition for identity validation and image recognition for medical diagnosis.

MDacne: computer vision technology fighting skin problems

MDacne is a mobile platform that uses AI and computer vision to help people with acne. Their mission is to disrupt the old physician-centric medical system and put the patient in the center using smart algorithms and computer vision to provide real-time mobile tools for people with health disorders. The company’s first mobile app, MDacne, seeks to help the 500 million people suffering from acne and provide them with an immediate self-diagnosis and personalized treatment plan.

Oded Harth, co-founder of MDacne: The idea was to really build what we see as the next stage in the evolution of dermatology.
The idea was to really build what we see as the next stage in the evolution of dermatology, which is a new type of app that you can simply take a picture of your skin or hair.
Oded Harth , MDacne CEO and co-founder

“The app analyzes the image immediately using artificial intelligence, and, based on that assessment, we deliver right to your doorstep a fully personalized treatment”, Oded said.

“First, because of the AI, the app actually can create new types of solutions, which is easier to use and accessible: everyone these days has a mobile phone with very high-resolution cameras, and also the new phones have a capability for artificial intelligence, so you can run it on the device itself,” MDacne CEO continued. “Second, as it is scalable and based on software, it is more affordable, because we do not need human dermatologists. Third, it is actually more effective, because AI is already at the level that it can find things in the skin that the human eye cannot find.”

MDacne has their own production line, manufacturing the treatment and thus guides the customer’s entire experience, from building the technology that analyzes the skin or hair to actually creating and customizing the treatments.

Challenges of building an AI

It was a lot of work: first of all, collecting data, labeling it, trying a lot of models, putting it on the device.
Oded Harth , MDacne CEO and co-founder

“Тhere were two kinds of main challenges and things we needed to achieve,” Oded Harth told.
“The first one was to achieve higher levels of accuracy. That meant trying lots of different models, playing with the data, and it took us a lot of time to really get it to be accurate enough. Thankfully, we have a dermatologist on the team, so we could do the labeling as well as the benchmarking and testing ourselves. That was quite a big challenge,” MDacne co-founder explained.

The company's first mobile app, MDacne, seeks to help the 500 million people suffering from acne and provide them with an immediate self-diagnosis and personalized treatment plan.

“Once we solved that one, we also wanted the AI to run on the device itself and not on the cloud, because, primarily, that gives us more privacy in the sense that the images don’t have to leave the phone if we don’t want them to.

But, more importantly, we wanted to run video recognition.

So the experience would be a bit fun, like a Snapchat filter, because when it’s on the cloud, the Internet speed is just not fast enough.

To get it on the device, we needed to find a model, shrink it, and put it on the device itself without affecting the accuracy; that was the second challenge,” Oded explained.

The biggest benefit of using computer vision algorithms is the scalability of the project. Getting to the accurate AI deep learning models took two years from launch, a journey that allows MDacne founders to recommend the best practices they used to succeed.

Recommendations for business leaders

First, think hard about how you are going to get your data in a scalable way, because you will need a lot of data.
Oded Harth , MDacne co-founder

“Second, think about how you’re going to get quality data, because, once you get the labeling part, you want to make sure the data is good, so not everything you just collected will need to be thrown out. That, I think, is a very important thing, because, at the end of the day, there are a lot of open-source models by the top universities and companies, but the data is something that can be your own kind of unique advantage, and, if you collect your own data, that’s something that can be just for you, and no one else,” MDacne co-founder recommended.

Hepta Airborne: computer vision for power lines maintenance

Hepta Airborne helps power line utilities to inspect their power grids with the help of drones to capture data and a specialized inspection platform called uBird to analyze it. Images are processed both manually and with the help of artificial intelligence. Computer vision algorithms detect issues in the power grid assets, enabling operators to fix them and ensuring that people have a more stable power supply.

We collect images: regular images, thermal images, LiDAR point clouds, all kinds of data. Now we have been experimenting and trying to implement AI technologies in order to do the inspections faster and more accurately.
German Bidzilja , Head of Product

As the field is rare and specific, ready-made visual datasets for machine learning are not big enough or may not even exist, threatening the possibility of this computer vision application. Consequently, the company (with MindTitan’s help) had to create its own dataset for image recognition training. Read the full version of the computer vision case study

Challenges of building an AI

German Bidzilja, Head of Product: Collaborating with MindTitan helped us to kick-start the AI development faster
We had to collect those datasets ourselves and have them in a good quality form for training the machine learning model.

German Bidzilja , Head of product

“The lack of openly available datasets was definitely one of the challenges there,” German said.

“A second challenge was that they also tried just picking the computer vision technology and understanding which kind of neural networks would work better and experimenting with that.”

Luckily, MindTitan supported us with several ideas to try, including how to gather better visual data. Together, we researched the usable kinds of data, the ways to label the images, and so on. MindTitan helped us in maintaining this project.
German Bidzilja , Head of product

Even though computer vision applications can sometimes be a challenge, using machine learning algorithms improves business processes. Having started small, Hepta Airborne keeps moving toward full automation of its processes. Yes, it takes time and effort, but, at the end of the day, it brings financial profit — which is the biggest benefit of computer vision technology for this company.

Recommendations for business leaders

Hepta Airborne provides power grid operators with an easily deployable end-to-end inspection suite and analysis tools.

“By far, the main criterion is the price. It has to be cheaper.

Speed, although it is good to have, is not that important, but, if it makes the service cheaper, that’s more,” German said.

We continue looking for different solutions, so the expectation from AI technologies is that we can process lots of images automatically without humans or, at least, with little human involvement.

German Bidzilja , Head of product

Neolook Solutions: saving lives with computer vision techniques

Neolook Solutions creates video-augmented services for neonatal and pediatric intensive care facilities to support family bonding, to encourage medical research via algorithms and AI, and to offer line of sight and alarm management and remote consultation for healthcare professionals. As we are talking about computer vision examples, we’ll focus on Neolook Solutions’ Screen2Screen Academic Extend. It serves academic research and the improvement of medical diagnostic methods with machine vision: saving lives with computer vision systems.

Marco D'Agata, Neolook Solutions founder: We use high-resolution video with long-form recording for AI development
We use high-resolution video with long-form recording for AI development.
Marco D'Agata , Neolook Solutions founder

“There is, for example, a scientifically proven connection between sleeping-generated movement patterns and the ones exhibited during awake time. Thus, we use AI to detect the movement patterns of the child,” Marco explained.

“Newborns movements are not controlled yet, can go all over the place. Parents know a baby can put their hand in their eye by accident.”

Neolook Solutions’ Screen2Screen Academic serves academic research and the improvement of medical diagnostic methods for physicians with machine vision.

“Such chaotic complexity is good because cognitive patterns are formed based on those motor patterns,” he mentioned.

“But, if a child moves asymmetrically or repetitively, then it can be an indication of dysfunction, which could be cognitive. If you have a video recording of a high-risk child’s movement patterns, you can send it to a doctor who can make a formal assessment.” Neolook Solutions founder said.

“We now create the AI to do the assessment via pattern recognition algorithms. The computer vision solution can monitor the movement patterns of the child and detect whether the child is at risk.”

“Another example is relationships observed between the frequency of the eyes blinking and a kind of slow onset lethargy that can be an indication of sepsis” Neolook Solutions founder said.

There are all kinds of key signs that may go unnoticed and when detected, can support nurses and doctors in their day-to-day work.
Marco D'Agata , Neolook Solutions founder

Faster diagnostics thanks to video recognition – this is the main benefit of computer vision technology for Neolook. Better results aren’t only due to data. Of course, using machine learning no doubt requires data, and as mentioned above, specific fields often do not have any publicly available datasets. Thus, Neolook Solutions created their own dataset too together with hospitals, but they stress the importance of good methods. If good analytics are there, some machine learning algorithms could work even without the big data to train on.

Challenges of building an AI

“I often use a story from pharmaceuticals, who do these big clinical double-blind randomized trials of 10,000 people, 20,000 people. That’s a regular, golden standard practice. So, lots of data and lots of power, but to get to statistical power, you don’t need lots of data–you need the right data. There’s this well-known example in the medical world. With only six mothers and a small sample of people, researchers proved that if the mothers used a certain medicine, their daughters got cancer,” Marco told.

So, you don’t need 10,000 people; you don’t need 5000, if you know what you’re doing. In this case, you need six; you need a very small number and you can do a perfect analysis.”
Marco D'Agata , Neolook Solutions founder

Recommendations for business leaders

Big data has become a buzzword, said our interlocutor. Moreover, it could seem that big data usage is an obligation for good computer vision work. However, praxis has shown that the quality of data reigns over its quantity. Thus, the main recommendation for business leaders is clear: choose your data sources wisely.

I would stay with the idea that it’s better to go small and specific and know what you’re doing than to go broad and big and lose your handle.
Marco D'Agata , Neolook Solutions founder


Many computer vision startups need an expert consultancy before implementing AI and computer vision. The main challenges are understanding what data they need, ways of collecting data, data labeling, and getting it to fit with the technical solution that can address their business needs. For example, MindTitan experts invest a lot of time in developing labeling strategies and educating clients on this topic.

In terms of the benefits of computer vision innovative solutions, we should mention the possibility of scaling the project, and time and money savings. The computer vision companies we talked to had one common challenge: getting the quality data for their computer vision systems. Thus, the main recommendations from the startup founders for business leaders who consider running computer vision companies are the following:

  1. Work towards generating a dataset aligned with the computer vision and business goals
  2. Understand that the quality of the data is as important or even more important than the quantity
  3. Know that the costs are more important than the speed of the process: businesses should benefit from computer vision and machine learning.


* This the edited version; an earlier version was unfortunately circulated not fully edited on Neolook Solutions’ part.

ai plan execution

Go back