Method and system for generating synthetic point cloud data using a generative model
AssigneesHuawei Technologies Co., Ltd., The Royal Institution for the Advancement of Learning/Mcgill University
InventorsLucas PAGÉ-CACCIA, Joelle PINEAU, Elmira AMIRLOO ABOLFATHI
Life of this patent
Estimating the term…
Abstract
Methods and systems for generating synthetic point cloud data are described. Projected 2D data grid is generated by projecting a 3D point cloud into a 2D grid, with rotation equivariance. A generative model is learned using the projected 2D data grid, wherein the generative model is implemented using flex-convolution and transpose flex convolution operations, for example in a generative adversarial network. The learned generative model is used to generate synthetic point clouds.
Claims20
A method comprising:
- obtaining a first batch of point clouds representing a three-dimensional (3D) environment;
- generating a projected two-dimensional (2D) data grid for each of the point clouds of the first batch of point clouds by projecting each point cloud of the first batch of point clouds into a projected 2D data grid with rotation equivariance;
- and
- during a training phase, learning a generative model which generates one or more batches of synthetic point clouds, the generative model being learned by providing the generative model with the projected 2D data grid for each of the point clouds of the first batch, wherein the generative model comprises flex-convolution operations and transpose flex-convolution operations.
A method comprising:
- obtaining a first batch of point clouds representing a three-dimensional (3D) environment;
- generating a projected two-dimensional (2D) data grid for each of the point clouds of the first batch of point clouds by projecting each point cloud of the first batch of point clouds into a projected 2D data grid with rotation equivariance;
- and
- during a training phase, learning a generative model which generates one or more batches of synthetic point clouds, the generative model being learned by providing the generative model with the first batch of point clouds, wherein the projected 2D data grids are used to identify nearest neighbors for performing flex-convolution operations and transpose flex-convolution operations during the learning.
A processing unit comprising:
- a processor;
- a memory storing computer-executable instructions which, when executed by the processor cause the processing unit to:
- obtain a first batch of point clouds representing a three-dimensional (3D) environment;
- generate a projected two-dimensional (2D) data grid for each of the point clouds of the first batch of point clouds by projecting each point cloud of the first batch of point clouds into a projected 2D data grid with rotation equivariance;
- and
- during a training phase, learn a generative model which generates one or more batches of synthetic point clouds, the generative model being learned by providing the generative model with the projected 2D data grid for each of the point clouds of the first batch, wherein the generative model comprises flex-convolution operations and transpose flex-convolution operations.
Dependencies are inferred from the claim text.