PhD Candidate at The University of Melbourne. Former Researcher in SIIT, Thailand and Lead Engineer in ListenField, Japan.
Our #OpenMutualLearning from #IGARSS2024 is out on #IEEE Xplore. #OpenStreetMap envisioned a worldwide #buildingfootprint dataset with community inputs for a while. Now, #Google and #Microsoft have created worldwide datasets with #CNNs trained on #remotesensing images. They all share a common problem - the datasets are always incomplete and updating them requires extra $$$. Our paper has devised a method to leverage the publicly available incomplete datasets and commercially available quasi-complete datasets to train multiple "Student networks" in a collaborative manner. The networks share each other's learnings during the training process to improve their performance. The sharing of learnings resulted in the improvement of the models trained on the incomplete datasets. The findings benefit anyone interested in creating large-scale inventories of #urbanenvironments and #digitaltwins. Paper: https://lnkd.in/g9YVeSXd I am thankful to my supervisors Jagannath Aryal and Abbas Rajabifard for their contributions to the research.
