Open-vocabulary object detections of inventory items

PreciBake is a computer vision tech company based in Munich, New York and Mumbai developing AI solutions for the gastronomy and baking industry. Our AI team is continuously working on developing and improving our machine learning algorithms for tasks such as image classification, object detection and tracking.

Computer vision based inventory tracking allows our restaurants and retail clients to measure changes in their food inventory in real-time. This can be used to dynamically schedule and optimize stock-ups of new inventory to ensure that we always have an optimal amount of fresh inventory available, without creating excessive inventory that could result in food waste.

Tracking inventory can be challenging since a typical inventory might contain hundreds or even thousands of unique items. As the inventory size grows, object detection becomes increasingly challenging and maintaining
high quality datasets becomes increasingly difficult. What makes it even more challenging is that new items could be introduced or their packaging could be changed which would occasionally require additional relabeling and retraining.

At the same time humans are able to easily recognize hundreds of thousands of different object categories without needing hundreds of training examples per category. In order to recognize new objects we only need a description of what the object looks like.

In this project we will study the challenging task of detecting an open set of objects merely based on the object description. Such a model could be used to detect a large number of different object categories including objects categories that it was not specifically trained for.