RENDERING DIGITAL VIRTUAL ENVIRONMENTS UTILIZING FULL PATH SPACE LEARNING

• US 20180089894A1
• Filed: 09/27/2016
• Published: 03/29/2018
• Est. Priority Date: 09/27/2016
• Status: Active Grant

• Current Assignee: Adobe Inc.
• Original Assignee: Adobe Systems Incorporated (Adobe Inc.)
• Inventors: Sun, Xin, Carr, Nathan, Qin, Hao
• Granted Patent: US 10,062,214 B2
• Time in Patent Office: Days
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06T 15/06 Ray-tracing
• CPC Class Codes: G06T 19/006 Mixed reality object pose determination, tracking or camera calibration for mixed reality G06T7/00
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/30244 Camera pose
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06T 7/50 Depth or shape recovery
• CPC Class Codes: G06T 7/70 Determining position or orientation of objects or cameras camera calibration G06T7/80
• CPC Class Codes: G06T 7/80 Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
• CPC Class Codes: H04N 23/71 Circuitry for evaluating the brightness variation

Abstract

The present disclosure includes methods and systems for rendering digital images of a virtual environment utilizing full path space learning. In particular, one or more embodiments of the disclosed systems and methods estimate a global light transport function based on sampled paths within a virtual environment. Moreover, in one or more embodiments, the disclosed systems and methods utilize the global light transport function to sample additional paths. Accordingly, the disclosed systems and methods can iteratively update an estimated global light transport function and utilize the estimated global light transport function to focus path sampling on regions of a virtual environment most likely to impact rendering a digital image of the virtual environment from a particular camera perspective.

Claims

1. In a digital medium environment, a computer-implemented method of efficiently rendering virtual environments utilizing full path space learning, comprising:

sampling a first plurality of paths in a virtual environment to determine point estimations of light transfer between a light source and a camera perspective using a uniform distribution;

approximating a global light transport function across full light paths between the light source and the camera perspective using the determined point estimations;

sampling a second plurality of paths in the virtual environment using a non-uniform distribution based on the approximated global light transport function across full light paths; and

creating a digital image of the virtual environment from the camera perspective based on the first plurality of paths and the second plurality of paths.

2. The method of claim 1, further comprising:

updating the approximated global light transport function based on the second plurality of paths;

repeatedly sampling additional paths using a distribution based on the updated global light transport function; and

repeatedly updating the updated global light transport function based on the sampled additional paths.

3. The method of claim 1, wherein sampling the first plurality of paths comprises sampling a plurality of paths originating from the light source and sampling a plurality of paths originating from the camera perspective.

4. The method of claim 3, wherein approximating the global light transport function comprises:

generating a first approximation of the global light transport function based on the plurality of paths originating from the light source; and

generating a second approximation of the global light transport function based on the plurality of paths originating from the camera perspective.

5. The method of claim 4, wherein:

sampling the second plurality of paths in the virtual environment based on the approximated global light transport function across full light paths comprises;

sampling an additional plurality of paths originating from the light source based on the first approximation of the global light transport function; and

sampling an additional plurality of paths originating from the camera perspective based on the second approximation of the global light transport function; and

creating the digital image of the virtual environment from the camera perspective based on the first plurality of paths and the second plurality of paths comprises;

creating the digital image based on the additional plurality of paths originating from the light source and the additional plurality of paths originating from the camera perspective.

6. The method of claim 1, wherein approximating the global light transport function comprises estimating the global light transport function by approximating the determined point estimations with a plurality of Gaussian functions.

7. The method of claim 6, wherein approximating the determined point estimations with the plurality of Gaussians comprises:

determining for the determined point estimations;

a first probability that a point estimation should be assigned to an existing Gaussian; and

a second probability that the point estimation should be assigned to a new Gaussian; and

based on which of the first probability or the second probability is larger;

assigning the determined point estimation to an existing Gaussian and updating a coefficient of the existing Gaussian;

or

fitting a new Gaussian to the determined point estimation.

8. The method of claim 6, wherein:

sampling the first plurality of paths in a virtual environment to determine point estimations of light transfer between the light source and the camera perspective comprises;

sampling a plurality of paths having a first number of segments; and

sampling a plurality of paths having a second number of segments; and

approximating the global light transport function with the plurality of Gaussian functions comprises;

approximating the global light transport function with a first plurality of Gaussian functions having a first number of dimensions corresponding to the first number of segments; and

approximating the global light transport function with a second plurality of Gaussian functions having a second number of dimensions corresponding to the second number of segments.

9. The method of claim 8, wherein estimating the global light transport function with the plurality of Gaussian functions comprises combining the first plurality of Gaussian functions having the first number of dimensions and the second plurality of Gaussian functions having the second number of dimensions utilizing a combination weighting.

10. The method of claim 8, wherein estimating the global light transport function with the plurality of Gaussian functions comprises distributing at least one of the first plurality of Gaussian functions having the first number of dimensions to a different dimension.

11. In a digital medium environment, a computer-implemented method of efficiently rendering virtual environments utilizing full path space learning, comprising:

sampling a first plurality of paths in a virtual environment with a light source and a camera perspective using a first distribution;

approximating a global light transport function across full light paths between the light source and the camera perspective based on the sampling of the first plurality of paths;

sampling a second plurality of paths in the virtual environment using a second distribution based on the approximated global light transport function across full light paths;

updating the approximated global light transport function across full light paths between the light source and the camera perspective based on the sampling of the second plurality of paths;

repeatedly sampling additional paths using distributions based on the updated global light transport function and repeatedly updating the updated global light transport function based on the sampled additional paths; and

creating a digital image of the virtual environment from the camera perspective based on the first plurality of paths, the second plurality of paths, and the additional paths.

12. The method of claim 11, wherein:

the approximated global light transport function defines probabilities that paths will transport light between the light source and the camera perspective; and

sampling the second plurality of paths comprises;

sampling the second plurality of paths based on the probabilities.

13. The method of claim 11, wherein:

sampling the second plurality of paths comprises sampling a plurality of paths originating from the light source and sampling a plurality of paths originating from the camera perspective;

updating the approximated global light transport function across light paths comprises;

updating a first approximated global light transport function based on the plurality of paths originating from the light source; and

updating a second approximated global light transport function based on the plurality of paths originating from the camera perspective; and

creating the digital image of the virtual environment from the camera perspective based on the approximated updated global model comprises;

creating the digital image based on the plurality of paths originating from the light source and the plurality of paths originating from the camera perspective.

14. The method of claim 11, wherein updating the global light transport function comprises estimating the global light transport function with a plurality of Gaussian functions by:

determining, based on the second plurality of paths, a number of initial Gaussian functions, the initial Gaussian functions having a position, and a variance; and

utilizing the number, the position, and the variance of the initial Gaussian functions to generate the plurality of Gaussian functions.

15. The method of claim 11, wherein:

sampling the second plurality of paths comprises;

sampling a first plurality of paths having a first number of segments; and

sampling a second plurality of paths having a second number of segments; and

updating the approximated global light transport function across full light paths comprises;

estimating the global light transport function with a first plurality of Gaussian functions having a first dimension corresponding to the first number of segments;

estimating the global light transport function with a second plurality of Gaussian functions having a second dimension corresponding to the second number of segments; and

combining the first plurality of Gaussian functions having the first dimension and the second plurality of Gaussian functions having the second dimension utilizing a combination weighting.

16. A system for efficiently rendering virtual environments utilizing full path space learning, comprising:

at least one processor; and

at least one non-transitory computer readable storage medium storing instructions thereon, that, when executed by the at least one processor, cause the system to;

sample a first plurality of paths in a virtual environment to determine point estimations of light transfer between a light source and a camera perspective;

approximate a global light transport function across full light paths between the light source and the camera perspective by;

fitting a first Gaussian to a first determined point estimation;

determining a first probability that a second determined point estimation should be assigned to the first Gaussian;

determining a second probability that the second determined point estimation should be assigned to a new Gaussian;

based on which of the first probability or the second probability is larger;

assigning the second determined point estimation to the first Gaussian and updating a coefficient of the first Gaussian;

or

fitting a second Gaussian to the second determined point estimation;

sample a second plurality of paths in the virtual environment using a distribution based on the approximated global light transport function across full light paths; and

render a digital image of the virtual environment from the camera perspective based on the first plurality of paths and the second plurality of paths.

17. The system of claim 16, further comprising instructions that, when executed by the at least one processor, cause the system to update the approximated global light transport function across full light paths between the light source and the camera perspective by:

determining new point estimations from the second plurality of paths; and

based on the new point estimations;

updating coefficients of existing Gaussians;

fitting new Gaussians to one or more of the new point estimations; and

removing an empty Gaussian.

18. The system of claim 17, wherein the instructions, when executed by the at least one processor, cause the system to fit a covariance of the new Gaussians to all of the determined point estimations and the new point estimations.

19. The system of claim 17, further comprising instructions that, when executed by the at least one processor, cause the system to:

repeatedly sample additional paths using distributions based on the updated global light transport function; and

repeatedly update the updated global light transport function based on the sampled additional paths.

20. The system of claim 18, further comprising instructions that, when executed by the at least one processor, cause the system to jointly fit Gaussians with different dimensionality by aligning low dimensional Gaussians to high dimensional Gaussians by extending rest dimensionality with uniform distribution.