Datum održavanja: srijeda, 10.4.2019. u 13:00 sati, prostorija O-403 (Vijećnica Odjela za informatiku)

Predavač: Goran Paulin

Naziv predavanja: Method for procedural generation of optimal image datasets

Abstract:

Creation and annotation of image datasets, designed for training machine learning models, is a very time consuming and costly process when it is based on a photo or video sources. It can be replaced by artificial data synthesis, providing 100% error-free segmentation by pixel-level classification of every object in the scene. Generation of such artificial data is currently based on building 3D scenes with already made 3D objects. We are proposing a method for artificial data synthesis based on a procedural generation of 3D objects and scenes along with the optimal set of parameters to efficiently produce image datasets for training models with good generalization.

Keywords: artificial data synthesis, image dataset, procedural generation, convolutional neural networks