US 12,273,545 B2
Task-driven machine learning-based representation and compression of point cloud geometry
Danillo Graziosi, Flagstaff, AZ (US); Alexandre Zaghetto, San Jose, CA (US); and Ali Tabatabai, Cupertino, CA (US)
Assigned to SONY GROUP CORPORATION, Tokyo (JP); and SONY CORPORATION OF AMERICA, Tokyo (JP)
Filed by SONY GROUP CORPORATION, Tokyo (JP); and Sony Corporation of America, New York, NY (US)
Filed on May 31, 2022, as Appl. No. 17/828,392.
Claims priority of provisional application 63/221,545, filed on Jul. 14, 2021.
Prior Publication US 2023/0025378 A1, Jan. 26, 2023
Int. Cl. H04N 19/42 (2014.01); G06N 3/082 (2023.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); H04N 19/597 (2014.01)
CPC H04N 19/42 (2014.11) [G06N 3/082 (2013.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); H04N 19/597 (2014.11)] 27 Claims
OG exemplary drawing
 
1. A method programmed in a non-transitory memory of a device comprising:
determining a task;
receiving a point cloud;
adjusting a neural network based on the task, including weighting specific variables depending on the task;
training the neural network with the point cloud;
compressing the neural network, including defining a function to indicate a probability of a point being occupied; and
sending the compressed neural network.