Robert is a machine learning engineer at Cloudonia, part of Megatrend group, where he is in charge of integrating and deploying ML models into production. Through college and previous projects, he mastered various fields, starting as a backend developer, then ventured into the world of machine learning, and finally learned the basics of DevOps and putting projects into production. Working on this project enabled him to consolidate all the acquired knowledge and to work on completely new challenges each day.
Computer Vision goes shopping – A System for Visual Recognition of Shelved Products
The project’s primary goal could roughly be placed in Retail Store Item Detection, but we are also targeting some additional information related to the products. The additional information should enable automated inspection of product placement, and price checking, but these tasks open a variety of specific technical problems. Fortunately, the essential tool to solve those problems is the same as for store item detection, so we detect them all: shelves, empty spaces, price tags, numerals of the prices, etc. Additional information on the product is also gathered by extending the product detection to detect important product parts, estimate product orientation, and category.
Unsurprisingly, deep neural networks have proven to be the right tool for all tasks, but some very specific modifications to the popular architectures are necessary. Combining many years of experience in the industry and innovative approaches, Megatrend and Ph.D. students at the Faculty of Electrical Engineering and Computing created the perfect team to tackle these problems.