Geometry encoder

Patent metadata

Field	Value
Publication number	US-10318891-B1
Application number	US-201816042712-A
Country	US
Kind code	B1
Filing date	Jul 23, 2018
Priority date	Jul 23, 2018
Publication date	Jun 11, 2019
Grant date	Jun 11, 2019

Abstract

A method includes receiving geometric data to be encoded, generating a signature for the geometric data based on the at least one property associated with the geometric data, enumerating a first set of options, enumerating a second set of options, encoding the geometric data using the first option and the second option, decoding the encoded geometric data, determining a performance associated with encoding the geometric data, determining a performance associated with decoding the encoded geometric data, and training a regressor based on the signature, the enumerated first option, the enumerated second option, the performance associated with encoding the geometric data and the performance associated with decoding the encoded geometric data.

First claim

What is claimed is:

1. A method comprising:
receiving geometric data to be encoded;
generating a signature for the geometric data based on the at least one property associated with the geometric data;
enumerating a first set of options;
enumerating a second set of options;
encoding the geometric data using the first option and the second option;
decoding the encoded geometric data;
determining a performance associated with encoding the geometric data;
determining a performance associated with decoding the encoded geometric data; and
training a regressor based on the signature, the enumerated first option, the enumerated second option, the performance associated with encoding the geometric data and the performance associated with decoding the encoded geometric data.

2. The method of claim 1 , wherein
the geometric data is mesh data,
the at least one property includes at least one of a number of vertices, a number of edges, and a number of triangles, and
the signature is based on the number of vertices, the number of edges, and the number of triangles in the mesh data.

3. The method of claim 1 , wherein
the geometric data is mesh data,
the at least one property includes a number of connected components for at least one attribute, and
the signature is based on the number of connected components for the at least one attribute in the mesh data.

4. The method of claim 1 , wherein
the geometric data is mesh data,
the at least one property includes a number of boundary edges for at least one attribute, and
the signature is based on the number of boundary edges for the at least one attribute in the mesh data.

5. The method of claim 1 , wherein
the geometric data is mesh data,
the at least one property includes an angle of a triangles corner, and
the signature is based on a statistical analysis of a histogram of the angles of triangle corners in the mesh data.

6. The method of claim 1 , wherein
the geometric data is mesh data,
the at least one property includes angles between triangles, and
the signature is based on a statistical analysis of a histogram of the angles between triangles in the mesh data.

7. The method of claim 1 , wherein
the geometric data is mesh data,
the at least one property includes vertex valences, and
the signature is based on a statistical analysis of a histogram of the vertex valences in the mesh data.

8. The method of claim 1 , wherein enumerating the second set of options includes enumerating all of the second set of options.

9. The method of claim 1 , wherein the first option includes a fixed option and an environmental option.

10. A method comprising:
receiving geometric data to be encoded;
determining at least one property associated with the geometric data;
generating a signature for the geometric data based on the at least one property;
receiving a first set of options;
enumerating a second set of options;
accessing a regressor based on the signature, the first set of options and the second set of options;
using the regressor to provide a performance estimate associated with each of the enumerated second set of options, the regressor including a plurality of performance estimates;
selecting a second option from the enumerated second set of options based on the plurality of performance estimates and a cost function; and
encoding the geometric data using the first set of options and the selected second option.

11. The method of claim 10 , wherein
the geometric data is mesh data,
the signature is based on a number of vertices, a number of edges, and a number of triangles in the mesh data, and
accessing the regressor uses a trained machine learning model based on one of a random forest model, a neural network model and a cluster analysis model.

12. The method of claim 10 , wherein
the geometric data is mesh data,
the signature is based on a number of connected components for at least one attribute in the mesh data, and
accessing the regressor uses a trained machine learning model based on one of a random forest model, a neural network model and a cluster analysis model.

13. The method of claim 10 , wherein
the geometric data is mesh data,
the signature is based on a number of boundary edges for at least one attribute in the mesh data, and
accessing the regressor uses a trained machine learning model based on one of a random forest model, a neural network model and a cluster analysis model.

14. The method of claim 10 , wherein
the geometric data is mesh data,
the signature is based on a histogram of an angle of a triangles corner in the mesh data, and
accessing the regressor uses a trained machine learning model based on one of a random forest model, a neural network model and a cluster analysis model.

15. The method of claim 10 , wherein
the geometric data is mesh data,
the signature is based on a histogram of angles between triangles in the mesh data, and
accessing the regressor uses a trained machine learning model based on one of a random forest model, a neural network model and a cluster analysis model.

16. The method of claim 10 , wherein
the geometric data is mesh data,
the signature is based on a histogram of vertex valences in the mesh data, and
accessing the regressor uses a trained machine learning model based on one of a random forest model, a neural network model and a cluster analysis model.

17. The method of claim 10 , wherein the first set of options include a fixed option and an environmental option.

18. The method of claim 10 , wherein
the first set of options include a fixed option and an environmental option,
the selecting of the second option is based on the fixed option and the environmental option, and
the encoding of the geometric data uses the fixed option.

19. The method of claim 10 , wherein the cost function is based on a performance associated with encoding the geometric data and a performance associated with decoding the encoded geometric data.

20. A non-transitory computer-readable storage medium having stored thereon computer executable program code which, when executed on a computer system, causes the computer system to perform steps comprising:
receiving geometric data to be encoded;
determining at least one property associated with the geometric data;
generating a signature for the geometric data based on the at least one property;
receiving a first set of options;
enumerating a second set of options;
accessing a regressor based on the signature and the first set of options;
using the regressor to provide a performance estimate associated with each of the enumerated second set of options, the regressor including a plurality of performance estimates;
selecting a second option from the enumerated second set of options based on the plurality of performance estimates and a cost function; and
encoding the geometric data using the first set of options and the selected second option.

Assignees

• Google Llc

Inventors

• Hemmer Michael
• Vytyaz Igor
• Makadia Ameesh
• Diez Leopoldo Taravilse

Classifications

• G06N3/08

  Learning methods · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06T9/002

  using neural networks · CPC title

• G06T9/00

  Image coding (bandwidth or redundancy reduction for static pictures H04N1/41; coding or decoding of static colour picture signals H04N1/64; methods or arrangements for coding, decoding, compressing or decompressing digital video signals H04N19/00) · CPC title

• G06T9/001 Primary

  Model-based coding, e.g. wire frame · CPC title