Recently published research from our academic founder Dr. Vladimir Krylov and Mr. Waqar Ahmad at the School of Mathematics and ADAPT Research Centre, DCU investigates new methods to optimise the means by which digital records of public infrastructure can be maintained to a high level of accuracy leveraging existing records in the process.

The research looks at the extraction of activation maps from convolutional neural networks trained for classification allows to automate the manual labour of human annotators in the presence of (historical) inventory records. This method dramatically reduces the time and resource-intensive human intervention required in the training process, and hence significantly facilitate the onboarding of new asset types into the street-level object detection pipeline.

Read the full paper here:

https://eurasip.org/Proceedings/Eusipco/Eusipco2024/pdfs/0000641.pdf

‍