Accepted article at the WSCG 2022
03.05.2022 -
Our article "Annotron: Semi-automatic Acquisition of Datasets for Retail Recognition" by Marco Filax, Tim Gonschorek and Frank Ortmeier has been accepted for publication.
The paper proposes an approach to reduce the manual annotation effort of labelers by reducing the time spent on repetitive, error-prone tasks. We evaluate the approach in the retail recognition domain, a fine-grained domain, which means that instance-level annotations are costly due to the density of products on the shelves. We refine an existing retrial product dataset and outperform the number of stock keeping units found previously. We labeled over 446,500 individual bounding boxes from 1188 different SKUs.