shares first-look on image recognition for retail analytics using deep learning shares first-look on image recognition for retail analytics using deep learning


On stage at GPU Technology Conference, Company revealed the initial results of joint case study with Nielsen on deep learning application for retail analytics and reference data management.

Robert Bogucki,’s Chief Science Officer, presented the first results of a joint project with Nielsen proving how state-of-the-art deep learning techniques can be applied to business, in this case, retail analytics. Bogucki was joined on stage by Alessandro Zolla, vice president of Technology at Nielsen.

Bogucki and Zolla explained how various information about a product - its category and ingredients, for example - can be retrieved automatically, solely from photographs of the packaging. The case study they presented at this year’s GTC involved identifying ingredient fields in photos of different kinds of FMCG products packaging. The solution was based on the interplay of region-based convolutional neural networks and NLP techniques.

Robert Bogucki described the project as “a great example how deep learning techniques can support core business processes, making them far more efficient. In retail market research, for example, it is crucial to extract product details. Using a mixture of visual and textual information, our algorithms overcame challenges such as bad image quality, different packaging shapes, and materials. The system achieved over 90% accuracy, which is a very good result, one that allows process automation and translates into measurable benefits.”

Alessandro Zolla commented, “This approach applied to CPG product labels, compared to simple optical character recognition, allows to retain the 'context' of specific region of interest. This makes the difference between 'extracting raw text' from the image versus 'understanding' image content allowing for a better accuracy and, consequently, higher quality results and higher automation rates. Nielsen is investing significantly in "smart" automation in the area of reference data which is one of the company key assets and the glue that enables internal and external data exchanges.”

GTC is held in Silicon Valley every year and this is not the first time’s data science team has been invited to share their accomplishments in deep learning. In 2016, Robert Bogucki also appeared there as a speaker with the facial recognition case study for right whales - the solutions delivered for NOAA in Kaggle’s global competition (the winning solution).

About delivers deep learning solutions and supports organizations in unlocking their data potential at all stages of data science maturity by helping: discover the power of deep learning and test it in their business, develop their data science capabilities and build end-to-end solutions, organize their data science process to increase its effectiveness. Neptune,’s newest technology, is the DevOps platform for data scientists designed to give data science teams the freedom to explore, while remaining organized and collaborative throughout the creative process. The second product Seahorse, a visual framework powered by Apache Spark, helps data scientists in rapid development of Spark is founded and managed by CodiLime — an IT consulting and software engineering company established by former Google, Facebook and Microsoft engineers.

Read our blog:
Try NeptuneGo! - an easily accessible public version of Neptune:



Download All

Deep Learning for retail analytics [GTC 2017].pdf

pdf | 3.83 MB


Intel brings data science to Polish universities with an event series powered by

Intel has invited to co-operate on the Intel Student Developer Program for AI,  which was designed to help students learn the practical side of machine learning. The program builds on the success Intel has achieved in the US.