公开(公告)号：US20230290135A1

公开(公告)日：2023-09-14

申请号：US18119770

申请日：2023-03-09

Applicant: NVIDIA Corporation

Inventor： Daquan Zhou ,  Zhiding Yu ,  Enze Xie ,  Anima Anandkumar ,  Chaowei Xiao ,  Jose Manuel Alvarez Lopez

IPC: G06V10/82 ,  G06V10/77 ,  G06V10/778 ,  G06V10/30

CPC classification number: G06V10/82 ,  G06V10/7715 ,  G06V10/778 ,  G06V10/30

Abstract: Apparatuses, systems, and techniques to generate a robust representation of an image. In at least one embodiment, input tokens of an input image are received, and an inference about the input image is generated based on a vision transformer (ViT) system comprising at least one self-attention module to perform token mixing and a channel self-attention module to perform channel processing.

公开(公告)号：US20230186428A1

公开(公告)日：2023-06-15

申请号：US18106348

申请日：2023-02-06

Applicant: NVIDIA Corporation

Inventor： Guilin Liu ,  Andrew Tao ,  Bryan Christopher Catanzaro ,  Ting-Chun Wang ,  Zhiding Yu ,  Shiqiu Liu ,  Fitsum Reda ,  Karan Sapra ,  Brandon Rowlett

IPC: G06T3/40 ,  G06T7/40 ,  G06N3/08 ,  G06V10/776 ,  G06V10/82 ,  G06V10/44 ,  G06V10/54

CPC classification number: G06T3/4038 ,  G06T3/4046 ,  G06T7/40 ,  G06N3/08 ,  G06V10/776 ,  G06V10/82 ,  G06V10/454 ,  G06V10/54 ,  G06T2207/20081 ,  G06T2207/20084

Abstract: Apparatuses, systems, and techniques for texture synthesis from small input textures in images using convolutional neural networks. In at least one embodiment, one or more convolutional layers are used in conjunction with one or more transposed convolution operations to generate a large textured output image from a small input textured image while preserving global features and texture, according to various novel techniques described herein.

公开(公告)号：US20220261593A1

公开(公告)日：2022-08-18

申请号：US17177068

申请日：2021-02-16

Applicant: NVIDIA Corporation

Inventor： Zhiding Yu ,  Shiyi Lan ,  Chris Choy ,  Subhashree Radhakrishnan ,  Guilin Liu ,  Yuke Zhu ,  Anima Anandkumar

IPC: G06K9/62 ,  G06T7/11 ,  G06K9/00 ,  G06N3/04 ,  G06N3/08

Abstract: Apparatuses, systems, and techniques to train one or more neural networks. In at least one embodiment, one or more neural networks are trained to perform segmentation tasks based at least in part on training data comprising bounding box annotations.

公开(公告)号：US20210064907A1

公开(公告)日：2021-03-04

申请号：US16998890

申请日：2020-08-20

Applicant: NVIDIA Corporation

Inventor： Xiaodong Yang ,  Yang Zou ,  Zhiding Yu ,  Jan Kautz

IPC: G06K9/44 ,  G06K9/46 ,  G06K9/62

Abstract: Object re-identification refers to a process by which images that contain an object of interest are retrieved from a set of images captured using disparate cameras or in disparate environments. Object re-identification has many useful applications, particularly as it is applied to people (e.g. person tracking). Current re-identification processes rely on convolutional neural networks (CNNs) that learn re-identification for a particular object class from labeled training data specific to a certain domain (e.g. environment), but that do not apply well in other domains. The present disclosure provides cross-domain disentanglement of id-related and id-unrelated factors. In particular, the disentanglement is performed using a labeled image set and an unlabeled image set, respectively captured from different domains but for a same object class. The identification-related features may then be used to train a neural network to perform re-identification of objects in that object class from images captured from the second domain.