Case study card:

Detecting product ingredients from images

Detecting product ingredients from images


eCommerce & Marketing

Business Challenge:
The Polish retailer owns thousands images of product labels. The task was to automatically find the location of the ingredients on packaging and transform this into text. The client was looking to offer this capability in a consumer facing app, to help with lifestyle and allergy based food choices.

Supplier solution:
They used a modified VGG16 convolutional neural network to find boxes containing ingredients information and as similar to ground truth boxes as possible. Extracted boxes were used in Google OCR to extract text data. After text data was extracted, they processed text using NLP techniques and further product categorisation was performed.

The algorithm was integrated into mobile application, so had to be highly efficient.

The solution processes images much faster than the manual alternative, helping in choosing right food to eat if somebody has allergies and used to check if there were product changes over time.




Confidential (Polish retail giant)