The point cloud contains all the data necessary to depict this statue. | Graphic: Alexiou Evangelos et al. (2020)

Point clouds provide immersive 3D experiences, e.g. in virtual reality, because they contain information about both the spatial position of objects and their colour and texture. But transmitting these huge data sets eats up resources. This is why Touradj Ebrahimi’s team at EPFL is trying to optimise the compression of point clouds by using artificial intelligence. After training with only 10,000 samples, their neural network already provides competitive results when compared to state-of-the-art compression such as MPEG. According to the EPFL researchers, additional training will help it to outperform these results.

E. Alexiou et al.: Towards neural network approaches for point cloud compression. SPIE Optical Engineering and Applications (2020)