Data Science: Computer Vision
Corporate Communications Director
This vlog is devoted to computer vision. Anastasiya Bordak, Lead Data Scientist, shares her experience of computer vision projects. Mark Hillary, a writer and analyst focused on technology, introduces the questions.
Anastasiya Bordak: “Computer vision can be found even in places where you do not expect it to be. For example, in farming. We developed a solution, where an agricultural company uses drones to track the condition of fields.”
- What is computer vision in your view?
– Computer vision (CV) is about digitizing information from photos or videos. As a result, CV can answer questions like 1) Are there any cats on the photo? 2) How many and what kind of cats are there? 3) What are the trajectories of the cats’ movements? And so on…
- What are the limitations that you have? If I show to you a photo of several dogs, would you be able to pick out my dog?
– Of course, there are viewpoints, scale, illumination, and other limitations in computer vision, but we have great expertise in this area and we know how to overcome them. Actually, we did it in our two solutions.
The first solution is about controlling the quality of product layouts and the distribution of goods among stores. This solution compares the rules of how products should be ideally laid out on the store shelves to the way they are really placed, analyzing the photos made by merchandisers. Besides, in this solution, we deploy new products from the start of their sales using just good quality studio photos. So customers can see how their new products are presented and distributed among shops.
Our second solution is a system for detecting and classifying people just by one photo in the database. We use this solution with a thermometer in our office to identify employees with high temperatures.
As you can see, we can find your dog with just one photo.
There are other challenges in CV, not just technical ones. These are the issues related to privacy and ethics of using face recognition systems. There is an open discussion about deep fakes and the explainability of models.
- You mentioned a couple of solutions there: retail or checking people’s temperature as they enter the office. Can it be used in all different industries? Are there any other kinds of solutions that you designed?
– In many-many areas. It can be found in obvious fields like tracking customers in a shop to evaluate cashier performance or for instance to simplify access control by classifying employees in a factory or detecting car numbers.
Computer vision can be found even in places where you do not expect it to be. For example, in farming. We developed a solution, where an agricultural company uses drones to track the condition of fields. The system quickly identifies field areas with a low number of shots.
Real-time information helps farmers plant on time and, as a result, they get higher yields in these areas.
If we are talking about mobile apps, masks in Instagram or Snapchat is also a great example of a CV task. The app detects key points on the face to make the mask look natural in real-time.
I would like to add that computer vision is not only about finding something in photos or classifying objects in photos. Enhancing photo quality or applying artistic effects to a photo to make it look as painted by Vincent van Gogh, for instance, is also a CV task.
- It sounds like it could improve my Instagram if it made everything look like a painting. It’s very innovative. You are using very high-tech solutions. Where do you see the future? What is going to be the new thing or development next year?
Improvement in computer vision depends not only on the evolution of engineering thinking in the field of machine learning. It mostly depends on hardware resources. Quantum computing is actively developing now and I believe it is going to be a game-changer, not only in machine learning and CV but in all industries. However, I don’t want to take a ruler and make simple linear regression on things that are happening right now and predict something like a Malthusian disaster.
This blog post is a part of a series of video discussions on data science. Please share your thoughts and offer your topics for future videos by leaving your comments or suggestions here.