公开(公告)号：US20240404003A1

公开(公告)日：2024-12-05

申请号：US18326437

申请日：2023-05-31

Applicant: QUALCOMM Incorporated

Inventor： Debasmit DAS ,  Hyojin PARK ,  Shubhankar Mangesh BORSE ,  Yu FU ,  Oleksandr BAILO ,  Mohsen GHAFOORIAN ,  Fatih Murat PORIKLI

IPC: G06T3/40

Abstract: Certain aspects of the present disclosure provide techniques for training and using an instance segmentation neural network to detect instances of a target object in an image. An example method generally includes generating, through an instance segmentation neural network, a first mask output from a first mask generation branch of the network. The method further includes generating, through the instance segmentation neural network, a second mask output from a second, parallel, mask generation branch of the network. The second mask output is typically of a lower resolution than the first mask output. The method further includes combining the first mask output and second mask output to generate a combined mask output. Based on the combined mask output, an output of the instance segmentation neural network is generated. One or more actions are taken based on the generated output.

公开(公告)号：US20250094781A1

公开(公告)日：2025-03-20

申请号：US18468873

申请日：2023-09-18

Applicant: QUALCOMM Incorporated

Inventor： Babak EHTESHAMI BEJNORDI ,  Gaurav KUMAR ,  Amelie Marie Estelle ROYER ,  Eduardo ESTEVES ,  Tijmen Pieter Frederik BLANKEVOORT ,  Mohsen GHAFOORIAN

IPC: G06N3/0464 ,  G06N3/084

Abstract: Systems and techniques are described herein for training and using multitask machine learning models. For example, a computing device can obtain training data for a first task in a layer in a neural network; perform, based on a determination from a first gating mechanism, the shared function on shared features of the training data using at least one shared channel to generate a shared feature map; perform, based on the determination from the first gating mechanism, the first task-specific function on first task-specific features of the training data using at least one first task-specific channel to generate a first task-specific feature map; generate an output for the first task-specific branch based on performing the shared function on the shared features and performing the first task-specific function on the first task-specific features; and update at least one parameter of the first gating mechanism based on the output.

公开(公告)号：US20250086995A1

公开(公告)日：2025-03-13

申请号：US18463068

申请日：2023-09-07

Applicant: QUALCOMM Incorporated

Inventor： Ching-Wei TSENG ,  Mohsen GHAFOORIAN ,  Nikhil KHULLAR ,  Yu FU ,  Shang-Lun LEE ,  Abhijeet BISAIN

IPC: G06V20/70 ,  G06T7/11

Abstract: Disclosed are systems and techniques for image processing. For example, a computing device can generate, using a multi-task model, a segmentation output and a normal output based on image(s) of a scene and a gravity vector for the scene. The computing device can learn semantic prediction(s) based on comparing the segmentation output to at least one ground truth semantic segmentation map. The computing device can also learn normal prediction(s) based on comparing the normal output to at least one ground truth normal map. The computing device can extract a semantics normal from the semantic prediction(s) and the normal prediction(s). The computing device can optimize a regularization loss based on the semantics normal and the gravity vector for the scene by learning gravity-normal regularization(s) for the scene. The computing device can determine final semantic labels for regions of the scene based on the gravity-normal regularization(s).

公开(公告)号：US20240135559A1

公开(公告)日：2024-04-25

申请号：US18153480

申请日：2023-01-12

Applicant: QUALCOMM Incorporated

Inventor： Georgi DIKOV ,  Mohsen GHAFOORIAN ,  Joris Johannes Lambertus VAN VUGT

IPC: G06T7/50 ,  G06T7/73

CPC classification number: G06T7/50 ,  G06T7/73 ,  G06T2207/20016 ,  G06T2207/20084

Abstract: Systems and techniques are provided for generating depth information from one or more images. For instance, a method can include obtaining an image of a scene and obtaining depth information associated with one or more objects in the scene. The method can include processing, using an encoder of a neural network model, the image and the depth information to generate a feature representation of the image and the depth information. The method can further include processing, using a decoder of the neural network model, the feature representation of the image and the depth information to generate a depth output corresponding to the image.