Motion-based image generation using one or more neural networks
AssigneeNVIDIA Corporation
InventorsTing-Chun Wang, Arun Mallya, Ming-Yu Liu, Abhijit Patait, Aurobinda Maharana, Vignesh Ungrapalli, Yogender Kumar Gupta, Vipul Parashar, Vishesh Lokras
Life of this patent
Estimating the term…
Abstract
Apparatuses, systems, and techniques are presented to generate image data. In at least one embodiment, one or more images are generated of one or more objects using one or more neural networks trained based, at least in part, upon one or more camera motions with respect to the one or more objects.
Claims20
A processor, comprising:
- one or more circuits to generate one or more images of one or more objects using one or more neural networks trained based, at least in part, upon training data including pairs of source images and target images, depicting the one or more objects, augmented to represent one or more camera motions with respect to the one or more objects, wherein a same set of one or more augmentations are applied to a first subset of the pairs of source images and target images, and for a second subset of the pairs of source images and target images, the source images are augmented differently than the target images.
A system, comprising:
- one or more processors to generate one or more images of one or more objects using one or more neural networks trained based, at least in part, upon training data including pairs of source images and target images, depicting the one or more objects, augmented to represent one or more camera motions with respect to the one or more objects, wherein a same set of one or more augmentations are applied to a first subset of the pairs of source images and target images, and for a second subset of the pairs of source images and target images, the source images are augmented differently than the target images.
A method, comprising:
- generating one or more images of one or more objects using one or more neural networks trained based, at least in part, upon training data including pairs of source images and target images, depicting the one or more objects, augmented to represent one or more camera motions with respect to the one or more objects, wherein a same set of one or more augmentations are applied to a first subset of the pairs of source images and target images, and for a second subset of the pairs of source images and target images, the source images are augmented differently than the target images.
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:
- generate one or more images of one or more objects using one or more neural networks trained based, at least in part, upon training data including pairs of source images and target images, depicting the one or more objects, augmented to represent one or more camera motions with respect to the one or more objects, wherein a same set of one or more augmentations are applied to a first subset of the pairs of source images and target images, and for a second subset of the pairs of source images and target images, the source images are augmented differently than the target images.
An image generation system, comprising:
- one or more processors to generate one or more images of one or more objects using one or more neural networks trained based, at least in part, upon training data including pairs of source images and target images, depicting the one or more objects, augmented to represent one or more camera motions with respect to the one or more objects, wherein a same set of one or more augmentations are applied to a first subset of the pairs of source images and target images, and for a second subset of the pairs of source images and target images, the source images are augmented differently than the target images;
- and
- memory for storing network parameters for the one or more neural networks.
Dependencies are inferred from the claim text.