RU 
EN 
CN 
ES 
The Samara University's Laboratory of Automated Systems for Scientific Research is currently working on a unique project – the neural network X-ray. It can immediately process a photograph of a hard-to-access part (e. g., hidden under a machine) into its exact 3D model on the screen with all its geometric features. The task that previously required expensive laser scanners and manual labor can now be completed using a simple RGB camera integratable into a smartphone.
Postgraduate student and engineer at the Laboratory Gennady Algashev is among the product’s developers. Despite science not being his childhood dream (he opted for IT as the profession), he didn’t mind when his life took an unexpected turn.
The sooner, the better
– What was your childhood like? Were there prerequisites for a scientific future?
– At school, I didn’t think about science. My parents are not scientists, either. My father works at Volgaburmash as the deputy head of the electrical shop, and my mother is in property accounting and inventory. After the 9th grade, I realized that I wanted to study at the ‘6th faculty’ of former Samara State Aerospace University, and become a programmer. That is, more or less, what I did. After my third year, I started working in an IT company and getting paid for it. A little earlier, I began to have thoughts that it would be great not to leave the university but stay there to teach. I wanted to make studying and students' lives better and more meaningful; to become a young practitioner who teaches and works in the industry, and therefore delivers up-to-date knowledge. At the university, one had to do research in order to be successful. I found myself surrounded by people who supported me in my first attempts at scientific research. That's how I came to do science; in a roundabout way, while pursuing other goals.
– Where did you receive your secondary education? How did school influence your life choices?
– I switched from one school to another. I received most of my secondary education at the 81th Samara School. Then I decided to switch to another school because I wanted to graduate prepared for the entrance exams. But my choice of a life journey was most notably influenced by the university and its professors.
– How was it to study at Samara University?
– I was an exemplary student and was involved in a lot of extracurricular activities. I was a group president, helped the dean's office and the entire faculty. Perhaps this also influenced my decision to stay. At that time, I had to learn a lot of things on my own and had no guidance on how I should act. Now, as a mentor for the young people, I want to become a person who I needed at that time; someone who would motivate them to go into research, to participate in contests and conferences, and who would reveal the opportunities available to students. In science, the sooner you start, the faster you can succeed, and possibly monetize your success.
From idea to implementation in seven months
– When did you first come up with the idea of a “neural network X-ray”?
– This is not my first research. At some point, I realized that I needed to work on my candidate degree thesis. About a year and a half ago, my supervisor, Aleksandr Viktorovich Kupriyanov, doctor in engineering science, and I began digging into this subject. We wanted to study the problems in computer vision and ways to solve them. This is how this topic emerged. It took us about seven months to progress from the first ideas to the first results. It's very fast indeed, and it's definitely a success!
– How can you describe the solution? What opportunities does it offer and what are its potential areas of application?
– I taught the neural networks to identify an object and find its key points, similar to how human joints are recognized on an X-ray – even if a part of the object is hidden, unevenly illuminated, or in shadow. The technology can be applied in various fields, but it is especially in demand in medicine and robotics. The majority of the existing solutions are slow, require expensive sensors, or simply fail to perform when the object in the image is partially obscured. The next step in this direction is to reconstruct the object's position in space. To achieve this, I taught the classic Perspective-n-Point (PnP) algorithm to ignore inaccessible or obscured points. Unlike standard solutions, this version of the algorithm works faster and more reliably even if only half of the object is visible.
– How exactly does artificial intelligence work in this case?
– I made sure to train the neural network properly. Instead of toilsome manual markup, I created a synthetic image generator. It takes a 3D model and “photographs” it from different angles, under different lighting, with glare, noise, and even with virtual obstacles partially obscuring the object. The program automatically marks the locations of boundaries and key points without human intervention. This enables quick training of the system on any objects by simply uploading the model into the generator. To ensure that the system works reliably with a wide variety of geometries, I proposed a unified approach to key points markup. For example, the same markup is applied to corners, hole centres or definitive features. This “language of dots” makes the neural network more robust and allows it to process even unfamiliar shapes without loss of accuracy. The result is a system that runs in real time – up to 82 frames per second on a single video card. Even if an object is 25–50 % obscure, the system accurately determines its position and orientation. No special sensors or scanners are needed – just a regular camera.
We are applied science lazybones
– What attracts you the most in science?
– Multiple opportunities and support available for young scientists. In fact, science is not always complicated. Sometimes it searches for a solution to a simple problem. When first-year students ask me what scientific theme they should opt for, I ask them about their favorite subject and its current problems. Solving this problem can become the goal of a scientific research. This is what science is about – we find a problem and try to solve it in a way that has not been used before.
– In your opinion, what distinguishes modern young scientists from the older generation?
– There is a certain gap, of course. Young scientists have impostor syndrome; we know that we understand science less fundamentally than our adult colleagues. But we are ready for bold experiments. Previously researchers avoided certain methods that seemed to them unethical or frivolous. Young scientists are lazybones; they are ready to solve problems in the simplest way and find solutions that more experienced scientists might disregard. And it works! Things shouldn’t be overcomplicated. In my opinion, we are more “applied” scientists than fundamentalists.
The Samara engineer has developed a solution using a regular RGB camera to photograph an object and obtain an accurate 3D model even if part of the object is hidden from view.
Source: samara.aif.ru
