公开(公告)号：US20220262101A1

公开(公告)日：2022-08-18

申请号：US17670737

申请日：2022-02-14

Applicant: CARNEGIE MELLON UNIVERSITY

Inventor： Marios Savvides ,  Zhiqiang Shen ,  Fangyi Chen ,  Han Zhang

IPC: G06V10/774

Abstract: Disclosed herein is a system and method for improving the accuracy of an object detector when trained with a dataset having a significant number of missing annotations. The method uses a novel Background Recalibration Loss (BRL) which adjusts the gradient direction according to its own activation to reduce the adverse effect of error signals by replacing the negative branch of the focal loss with a mirror of the positive branch when the activation is below a confusion threshold.

公开(公告)号：US12266156B2

公开(公告)日：2025-04-01

申请号：US17670737

申请日：2022-02-14

Applicant: CARNEGIE MELLON UNIVERSITY

Inventor： Marios Savvides ,  Zhiqiang Shen ,  Fangyi Chen ,  Han Zhang

IPC: G06V10/774

Abstract: Disclosed herein is a system and method for improving the accuracy of an object detector when trained with a dataset having a significant number of missing annotations. The method uses a novel Background Recalibration Loss (BRL) which adjusts the gradient direction according to its own activation to reduce the adverse effect of error signals by replacing the negative branch of the focal loss with a mirror of the positive branch when the activation is below a confusion threshold.

公开(公告)号：US11954175B2

公开(公告)日：2024-04-09

申请号：US17386879

申请日：2021-07-28

Applicant: CARNEGIE MELLON UNIVERSITY

Inventor： Fangyi Chen ,  Chenchen Zhu ,  Zhiqiang Shen ,  Han Zhang ,  Marios Savvides

IPC: G06F18/214 ,  G06F18/213 ,  G06F18/2431 ,  G06N5/04 ,  G06N20/00

CPC classification number: G06F18/2148 ,  G06F18/213 ,  G06F18/2431 ,  G06N5/04 ,  G06N20/00

Abstract: Disclosed herein is an improvement to prior art feature pyramids for general object detection that inserts a simple norm calibration (NC) operation between the feature pyramids and detection head to alleviate and balance the norm bias caused by feature pyramid network (FPN) and which leverages an enhanced multi-feature selective strategy (MS) during training to assign the ground-truth to one or more levels of the feature pyramid.

公开(公告)号：US20240046621A1

公开(公告)日：2024-02-08

申请号：US18491059

申请日：2023-10-20

Applicant: CARNEGIE MELLON UNIVERSITY

Inventor： Marios Savvides ,  Fangyi Chen ,  Han Zhang ,  ChenChen Zhu

IPC: G06V10/774 ,  G06V10/77 ,  G06V10/82 ,  G06V10/766 ,  G06V10/764 ,  G06V10/776 ,  G06V30/18 ,  G06V30/19

CPC classification number: G06V10/774 ,  G06V10/7715 ,  G06V10/82 ,  G06V10/766 ,  G06V10/764 ,  G06V10/776 ,  G06V30/1801 ,  G06V30/19093 ,  G06V30/19147

Abstract: Disclosed herein are designs for two baselines to detect products in a retail setting. A novel detector, referred to herein as RetailDet, detects quadrilateral products. To match products using visual texts on 2D space, text features are encoded with spatial positional encoding and the Hungarian Algorithm that calculates optimal assignment plans between varying text sequences is used.

公开(公告)号：US20240104761A1

公开(公告)日：2024-03-28

申请号：US18272754

申请日：2022-03-09

Applicant: CARNEGIE MELLON UNIVERSITY

Inventor： Marios Savvides ,  Chenchen Zhu ,  Fangyi Chen ,  Han Zhang

IPC: G06T7/70 ,  G06V10/25 ,  G06V10/44 ,  G06V10/764

CPC classification number: G06T7/70 ,  G06V10/25 ,  G06V10/44 ,  G06V10/764 ,  G06V2201/07

Abstract: Disclosed herein is a system and method for generating quadrilateral bonding boxes which tightly cover the most representative faces of retail products having arbitrary poses. The quadrilateral boxes do not include unnecessary background information or miss parts of the objects, as would the axis-aligned bounding boxes produced by prior art detectors. A simple projection transformation can correct the pose of products for downstream tasks.

公开(公告)号：US20240054764A1

公开(公告)日：2024-02-15

申请号：US18267543

申请日：2022-02-04

Applicant: CARNEGIE MELLON UNIVERSITY

Inventor： Ran Tao ,  Marios Savvides ,  Han Zhang

IPC: G06V10/771

CPC classification number: G06V10/771

Abstract: Disclosed herein is a methodology implementing feature map-level data augmentation in a feature map. Two or more units in the feature map are selected and the values of locations in the two or more units are swapped among the two or more units. Value perturbations applied around local units in the feature map implicitly lead to an unused data augmentation at the image level.