Generating voxel representations using one or more neural networks
Apparatuses, systems, and techniques are presented to generate digital content. In at least one embodiment, one or more neural networks are used to generate a three-dimensional voxel representation of a scene based, at least in part, upon a plurality of two-dimensional images of the scene.
1. A processor, comprising:
one or more circuits to cause one or more neural networks to generate one or more hierarchical encodings each representing different types of scene information.
2. The processor of claim 1 , wherein the one or more circuits are further to cause a neural network to generate a three-dimensional voxel latent representation of a scene using a plurality of two-dimensional images of the scene and one or more parameters of one or more cameras used to capture the plurality of two-dimensional images, wherein the one or more hierarchical encodings are based on the three-dimensional voxel latent representation.
3. The processor of claim 2 , wherein the neural network is an autoencoder trained to generate the three-dimensional voxel latent representation based, at least in part, upon a density voxel representation and a feature voxel representation for this scene, generated using the plurality of two-dimensional images and the one or more parameters of the one or more cameras.
4. The processor of claim 2 , wherein the one or more neural networks used to generate one or more hierarchical encodings includes a hierarchical set of latent encoders to decompose the three-dimensional voxel latent representation of the scene into a set of smaller latent representations of the scene for use in learning a generative model.
5. The processor of claim 4 , wherein the generative model includes at least one diffusion neural network or a generative adversarial network.
6. The processor of claim 2 , wherein the neural network is trained to minimize a reconstruction loss and an adversarial loss in a three-dimensional space.
7. A system, comprising:
one or more processors to cause one or more neural networks to generate one or more hierarchical encodings each representing different types of scene information.
8. The system of claim 7 , wherein the one or more processors are further to cause a neural network to generate a three-dimensional voxel latent representation of a scene using a plurality of two-dimensional images of the scene and one or more parameters of one or more cameras used to capture the plurality of two-dimensional images, wherein the one or more hierarchical encodings are generated from the three-dimensional voxel latent representation.
9. The system of claim 8 , wherein the neural network is an autoencoder trained to generate the three-dimensional voxel latent representation based, at least in part, upon a density voxel representation and a feature voxel representation for this scene, generated using the plurality of two-dimensional images and the one or more parameters of the one or more cameras.
10. The system of claim 8 , wherein the one or more neural networks used to generate one or more hierarchical encodings includes a hierarchical set of latent encoders to decompose the three-dimensional voxel latent representation of the scene into a set of smaller latent representations of the scene for use in learning a generative model.
11. The system of claim 10 , wherein the generative model includes at least one diffusion neural network or a generative adversarial network.
12. The system of claim 8 , wherein the neural network is trained to minimize a reconstruction loss and an adversarial loss in a three-dimensional space.
13. A method, comprising:
causing, using one or more processors, one or more neural networks to generate one or more hierarchical encodings each representing different types of scene information.
14. The method of claim 13 , further comprising:
causing a neural network to generate a three-dimensional voxel latent representation of a scene using a plurality of two-dimensional images of the scene and one or more parameters of one or more cameras used to capture the plurality of two-dimensional images, wherein the one or more hierarchical encodings are generated from the three-dimensional voxel latent representation.
15. The method of claim 14 , wherein the neural network is an autoencoder trained to generate the three-dimensional voxel latent representation based, at least in part, upon a density voxel representation and a feature voxel representation for this scene, generated using the plurality of two-dimensional images and the one or more parameters of the one or more cameras.
16. The method of claim 14 , wherein the one or more neural networks used to generate one or more hierarchical encodings includes
a hierarchical set of latent encoders to decompose the three-dimensional voxel latent representation of the scene into a set of smaller latent representations of the scene for use in learning a generative model.
17. The method of claim 16 , wherein the generative model includes at least one diffusion neural network or a generative adversarial network.
18. The method of claim 14 , wherein the neural network is trained to minimize a reconstruction loss and an adversarial loss in a three-dimensional space.
19. A non-transitory machine-readable medium having stored thereon a set of instructions, which if performed by one or more processors, cause the one or more processors to at least:
cause one or more neural networks to generate one or more hierarchical encodings each representing different types of scene information.
20. The non-transitory machine-readable medium of claim 19 , wherein the instructions, if performed, are further cause the one or more processors to cause a neural network to generate a three-dimensional voxel latent representation of a scene using a plurality of two-dimensional images of the scene and one or more parameters of one or more cameras used to capture the plurality of two-dimensional images, wherein the one or more hierarchical encodings are generated from the three-dimensional voxel latent representation.
21. The non-transitory machine-readable medium of claim 20 , wherein the neural network is an autoencoder trained to generate the three-dimensional voxel latent representation based, at least in part, upon a density voxel representation and a feature voxel representation for this scene, generated using the plurality of two-dimensional images and the one or more parameters of the one or more cameras.
22. The non-transitory machine-readable medium of claim 20 , wherein the one or more neural networks used to generate one or more hierarchical encodings includes
a hierarchical set of latent encoders to decompose the three-dimensional voxel latent representation of the scene into a set of smaller latent representations of the scene for use in learning a generative model.
23. The non-transitory machine-readable medium of claim 22 , wherein the generative model includes at least one diffusion neural network or a generative adversarial network.
24. The non-transitory machine-readable medium of claim 20 , wherein the neural network is trained to minimize a reconstruction loss and an adversarial loss in a three-dimensional space.
25. A scene generation system, comprising:
one or more processors to cause one or more neural networks to generate one or more hierarchical encodings each representing different types of scene information; and
memory for storing network parameters for the one or more neural networks.
26. The scene generation system of claim 25 , wherein the one or more processors are further to cause a neural network to generate a three-dimensional voxel latent representation of a scene using a plurality of two-dimensional images of the scene and one or more parameters of one or more cameras used to capture the plurality of two-dimensional images, wherein the one or more hierarchical encodings are generated from the three-dimensional voxel latent representation.
27. The scene generation system of claim 26 , wherein the neural network is an autoencoder trained to generate the three-dimensional voxel latent representation based, at least in part, upon a density voxel representation and a feature voxel representation for this scene, generated using the plurality of two-dimensional images and the one or more parameters of the one or more cameras.
28. The scene generation system of claim 26 , wherein the one or more neural networks used to generate one or more hierarchical encodings includes
a hierarchical set of latent encoders to decompose the three-dimensional voxel latent representation of the scene into a set of smaller latent representations of the scene for use in learning a generative model.
29. The scene generation system of claim 28 , wherein the generative model includes at least one diffusion neural network or a generative adversarial network.
30. The scene generation system of claim 26 , wherein the neural network is trained to minimize a reconstruction loss and an adversarial loss in a three-dimensional space.