Method and system for synthesizing three-dimensional data

1. A method for synthesizing three-dimensional (3D) data representing a plurality of possible scenarios from initial 3D data representing a given scenario, the method comprising:

clustering, by a data synthesis device, initial 3D data to identify one or more regions of interest (ROIs), wherein the initial 3D data represent a given scenario;

generating, by the data synthesis device, input specific noise data based on the one or more ROIs by an iterative process using Gaussian mixture model; and

iteratively synthesizing, by the data synthesis device, 3D data based on the one or more ROIs and the input specific noise data using generative adversarial networks (GANs) to generate final synthesized 3D data, wherein the final synthesized 3D data represent a plurality of possible scenarios and are affine transforms of the initial 3D data.

2. The method of claim 1 , wherein the initial 3D data are point cloud data acquired by a light detection and ranging (LiDAR) sensor.

3. The method of claim 1 , wherein the initial 3D data are high definition data.

4. The method of claim 3 , further comprising down sampling the final synthesized 3D data using one or more voxel grid filters.

5. The method of claim 1 , further comprising training an artificial intelligence (AI) model with the final synthesized 3D data.

6. The method of claim 5 , wherein the AI model is trained for object classification and localization for use in autonomous navigation application.

7. The method of claim 5 , further comprising:

determining Eigen values for the final synthesized 3D data using singular value decomposition; and

computing embedding vector for the final synthesized 3D data based on the Eigen values.

8. The method of claim 1 , wherein generating the input specific noise data comprises:

computing multi-variate Gaussian distribution of the one or more ROIs based on a set of feature vectors for the one or more ROIs; and

deriving prior probability for the multi-variate Gaussian distribution of the one or more ROIs based on a set of classes for the one or more ROIs and the set of feature vectors for the one or more ROIs.

9. The method of claim 1 , wherein generating the input specific noise data further comprises iteratively back-propagating the 3D data to generate updated input specific noise data.

10. A system for synthesizing three-dimensional (3D) data representing a plurality of possible scenarios from initial 3D data representing a given scenario, the system comprising:

a data synthesis device comprising of at least one processor and a computer-readable medium storing instruction that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:

clustering initial 3D data to identify one or more regions of interest (ROIs), wherein the initial 3D data represent a given scenario;

generating input specific noise data based on the one or more ROIs by an iterative process using Gaussian mixture model; and

iteratively synthesizing 3D data based on the one or more ROIs and the input specific noise data using generative adversarial networks (GANs) to generate final synthesized 3D data, wherein the final synthesized 3D data represent a plurality of possible scenarios and are affine transforms of the initial 3D data.

11. The system of claim 10 , wherein the initial 3D data are point cloud data acquired by a light detection and ranging (LiDAR) sensor.

12. The system of claim 10 , wherein the initial 3D data are high definition data.

13. The system of claim 12 , wherein the operations further comprise down sampling the final synthesized 3D data using one or more voxel grid filters.

14. The system of claim 10 , wherein the operations further comprise training an artificial intelligence (AI) model with the final synthesized 3D data.

15. The system of claim 14 , wherein the operations further comprise:

determining Eigen values for the final synthesized 3D data using singular value decomposition; and

computing embedding vector for the final synthesized 3D data based on the Eigen values.

16. The system of claim 10 , wherein generating the input specific noise data comprises:

computing multi-variate Gaussian distribution of the one or more ROIs based on a set of feature vectors for the one or more ROIs; and

deriving prior probability for the multi-variate Gaussian distribution of the one or more ROIs based on a set of classes for the one or more ROIs and the set of feature vectors for the one or more ROIs.

17. The system of claim 10 , wherein generating the input specific noise data comprises iteratively back-propagating the 3D data to generate updated input specific noise data.

18. A non-transitory computer-readable medium storing computer-executable instruction for:

clustering initial 3D data to identify one or more regions of interest (ROIs), wherein the initial 3D data represent a given scenario;

generating input specific noise data based on the one or more ROIs by an iterative process using Gaussian mixture model; and

iteratively synthesizing 3D data based on the one or more ROIs and the input specific noise data using generative adversarial networks (GANs) to generate final synthesized 3D data, wherein the final synthesized 3D data represent a plurality of possible scenarios and are affine transforms of the initial 3D data.

19. The non-transitory computer-readable medium of claim 18 , wherein generating the input specific noise data comprises:

computing multi-variate Gaussian distribution of the one or more ROIs based on a set of feature vectors for the one or more ROIs; and

deriving prior probability for the multi-variate Gaussian distribution of the one or more ROIs based on a set of classes for the one or more ROIs and the set of feature vectors for the one or more ROIs.

20. The non-transitory computer-readable medium of claim 18 , wherein generating the input specific noise data comprises iteratively back-propagating the 3D data to generate updated input specific noise data.