Magic Leap, Inc.

Nvidia Corporation

Jianyuan Min

Xiaolin Wei

David Geoffrey Molyneaux

Yifu Zhang

Zhongle Wu

Frank Thomas Steinbrücker

Rochelle Pereira

Vukasin Milovanovic

Joy D'Souza

Jan Kautz

Akarsh Umesh Zingade

Shuye Han

Mayoore Selvarasa Jaiswal

Systems and methods are disclosed related to synthetic bracketing for exposure correction. A deep learning based method and system produces a set of differently exposed images from a single input image. The images in the set may be combined to produce an output image with improved global and local exposure compared with the input image. An image encoder applies learned parameters to each input image to generate a set of image features including local exposure estimates for each of two or more regions of the input image and a low resolution latent representation of the input image. A decoder receives the local exposure estimates, the latent representation, and target enhancements that are processed to generate synthesized transformations. When applied to the input image, the synthesized transformations produce the set of transformed images. Each transformed image is a version of the input image synthesized to correspond to a respective target enhancement.

Synthetic bracketing for exposure correction

In various examples, image data may be received that represents an image. Corner detection may be used to identify pixels that may be candidate corner points. The image data may be converted from a higher dimensional color space to a converted image in a lower dimensional color space, and boundaries may be identified within the converted image. A set of the candidate corner points may be determined that are within a threshold distance to one of the boundaries, and the set of the candidate corner points may be analyzed to determine a subset of the candidate corner points representative of corners of polygons. Using the subset of the candidate corner points, one or more polygons may be identified, and a filter may be applied to the polygons to identify a polygon as corresponding to a fiducial marker boundary of a fiducial marker.

Graphical fiducial marker identification

A method to culling parts of a 3D reconstruction volume is provided. The method makes available to a wide variety of mobile XR applications fresh, accurate and comprehensive 3D reconstruction data with low usage of computational resources and storage spaces. The method includes culling parts of the 3D reconstruction volume against a depth image. The depth image has a plurality of pixels, each of which represents a distance to a surface in a scene. In some embodiments, the method includes culling parts of the 3D reconstruction volume against a frustum. The frustum is derived from a field of view of an image sensor, from which image data to create the 3D reconstruction is obtained.

Viewpoint dependent brick selection for fast volumetric reconstruction

Systems, processes, and techniques to automatically detect and enlarge a speaking one of plurality of participants on one side of a video conference. In at least one embodiment, the speaking participant is identified using one or more heuristics and/or one or more neural networks.

Machine learning techniques for enhancing video conferencing applications

An augmented reality/mixed reality system that provides a more immersive user experience. That experience is provided with increased speed of update for occlusion data by using depth sensor data augmented with lower-level reconstruction data. When operating in real-time dynamic environments, changes in the physical world can be reflected quickly in the occlusion data. Occlusion rendering using live depth data augmented with lower-level 3D reconstruction data, such as a raycast point cloud, can greatly reduce the latency for visual occlusion processing. Generating occlusion data in this way may provide faster operation of an XR system using less computing resources and enabling the system to be packaged in a battery operated wearable device.

Method of occlusion rendering using raycast and live depth

A method to efficiently update and manage outputs of real time or offline 3D reconstruction and scanning in a mobile device having limited resource and connection to the Internet is provided. The method makes available to a wide variety of mobile XR applications fresh, accurate and comprehensive 3D reconstruction data, in either single user applications or multi-user applications sharing and updating the same 3D reconstruction data. The method includes a block-based 3D data representation that allows local update and maintains neighbor consistency at the same time, and a multi-layer caching mechanism that retrieves, prefetches, and stores 3D data efficiently for XR applications.

Caching and updating of dense 3D reconstruction data

A method of operating a computing system to generate a model of an environment represented by a mesh is provided. The method allows to update 3D meshes to client applications in real time with low latency to support on the fly environment changes. The method provides 3D meshes adaptive to different levels of simplification requested by various client applications. The method provides local update, for example, updating the mesh parts that are changed since last update. The method also provides 3D meshes with planarized surfaces to support robust physics simulations. The method includes segmenting a 3D mesh into mesh blocks. The method also includes performing a multi-stage simplification on selected mesh blocks. The multi-stage simplification includes a pre-simplification operation, a planarization operation, and a post-simplification operation.

Multi-stage block mesh simplification

A method of merging 3D meshes includes receiving a first mesh and a second mesh; performing spatial alignment to register the first mesh and the second mesh in a common world coordinate system; performing mesh clipping on the first mesh and the second mesh to remove redundant mesh vertices; performing geometry refinement around a clipping seam to close up mesh concatenation holes created by mesh clipping; and performing texture blending in regions adjacent the clipping seam to obtain a merged mesh.

Growing community of inventors

Santa Clara, CA, United States of America

Jianyuan Min