Brussels / 2 & 3 February 2019


Improve OSM data quality with DeepLearning

Quality Analysis on a wide dataset, is always a gageure. And automatic semantic segmentation from imagery is still an open subject since decades. But. Nowadays with latests DeepLearning techniques, we can use new kind of techniques, to easily extract patterns from our dataset, and therefore help humans to be that more efficient to take the right decision (think filtering).

Two main points on this presentation: - How to produce high quality results, while you start with noisy/creapy/real world data ? - How to predict, at scale without huge hardware infrastructure ? as the ecosystem to do so.


Olivier Courtin