ROSETTA is a complete, end-to-end trainable neural network that can perform multiple objects detection in real time. The goal is to create a machine learning module that can be used to detect objects picked up by customers from refrigerators using images from a mounted camera. This will allow for automatic, real-time bookkeeping of a unit’s product catalog.
Since this module is needed for widespread deployment, it has to run on an easy-to-install, low-power device, like the Jetson Nano 4GB, without any help from the cloud. Even though object detection is a task that machine learning has excelled at, there are many technical challenges for this system. The hands of the shoppers will always be within view and potentially abstract the grabbed object. Images can be blurry. The background can be dynamic and so on. To overcome these challenges, novel methods will be used based on deep learning networks.
Usage of Standards for data interoperability:
- Open data formats (JSON, CSV)
- EU (eIDAS) date format
- REST API under Open API specification
- GDPR data processor requirements
- General guidelines of DISC (Data Interoperability Standards Consortium) and European Interoperability Framework (EIF)