A9.com, Inc.

Facebook, Inc.

Meta Platforms Technologies, LLC

Shanghsuan Tsai

Colin Jon Taylor

Mehmet Nejat Tek

Ismet Zeki Yalniz

Peter Vajda

Elad Joseph

Zijian He

Priyam Chatterjee

Evyatar Bluzer

Dmitrii Marin

Ezri Joseph Sonn

A method includes a computing system associated with a first artificial-reality device accessing sensor data of a real environment, generating a first plurality of feature descriptors for the real environment, and generating a second plurality of feature descriptors for the real environment based on the first plurality of feature descriptors, wherein the second plurality of feature descriptors have lower detail than the first plurality of feature descriptors. The computing system may further cause the first plurality of feature descriptors to be transmitted using a wireless connection, wherein the transmitted first plurality of feature descriptors are configured to be converted into a third plurality of feature descriptors different from the second plurality of feature descriptors. The third plurality of feature descriptors are configured to be used by a second artificial-reality device for tracking the real environment. The computing system tracks the real environment using the second plurality of feature descriptors.

Methods and systems to allow three-dimensional map sharing between heterogeneous computing systems, cross-localization, and sharing content in three-dimensional space

Techniques are described for content-adaptive downsampling of digital images and videos for computer vision operations, such as semantic segmentation. A computer vision system comprises a memory, one or more processors operably coupled to the memory and a downsampling module configured for execution by the one or more processors to perform, based on a non-uniform sampling model trained to predict content-aware sampling parameters, downsampling input image data to generate downsampled image data. A segmentation module is configured for execution by the one or more processors to perform segmentation on the downsampled image to produce a segmentation result, such as a feature map that assigns pixels of the downsampled image data to object classes. An upsampling module is configured for execution by the one or more processors to perform upsampling according to the segmentation result to produce upsampled image data.

Content-adaptive non-uniform image downsampling using predictive auxiliary convolutional neural network

Object identification through image matching can utilize ratio and other data to accurately identify objects having relatively few feature points otherwise useful for identifying objects. An initial image analysis attempts to locate a 'scalar' in the image, such as may include a label, text, icon, or other identifier that can help to narrow a classification of the search, as well as to provide a frame of reference for relative measurements obtained from the image. By comparing the ratios of dimensions of the scalar with other dimensions of the object, it is possible to discriminate between objects containing that scalar in a way that is relatively robust to changes in viewpoint. A ratio signature can be generated for an object for use in matching, while in other embodiments a classification can identify priority ratios that can be used to more accurately identify objects in that classification.

Image match for featureless objects

Growing community of inventors

Sunnyvale, CA, United States of America

Shanghsuan Tsai