The patent badge is an abbreviated version of the USPTO patent document. The patent badge does contain a link to the full patent document.
The patent badge is an abbreviated version of the USPTO patent document. The patent badge covers the following: Patent number, Date patent was issued, Date patent was filed, Title of the patent, Applicant, Inventor, Assignee, Attorney firm, Primary examiner, Assistant examiner, CPCs, and Abstract. The patent badge does contain a link to the full patent document (in Adobe Acrobat format, aka pdf). To download or print any patent click here.
Patent No.:
Date of Patent:
May. 26, 2026
Filed:
Jan. 12, 2024
The Regents of the University of Michigan, Ann Arbor, MI (US);
Hitachi Astemo Americas, Inc., Farmington Hills, MI (US);
Jason Mars, Ann Arbor, MI (US);
Zongyu Chen, Ann Arbor, MI (US);
Yiping Kang, Ann Arbor, MI (US);
Roland Daynauth, Ann Arbor, MI (US);
Arima Hidetoshi, Farmington Hills, MI (US);
Paul Liu, Farmington Hills, MI (US);
The Regents of The University of Michigan, Ann Arbor, MI (US);
Astemo Americas, Inc., Harrodsburg, KY (US);
Abstract
Currently, objection detection models require a large amount of training cycles on even the most powerful hardware, fundamentally limiting exploration of the model design space. This work proposes interchanging the constituent components of objection detection models without retraining to quickly build and evaluate new model designs. A typical objection detection model consists of multiple constituent components (submodels) and the conventional wisdom is to train all sub-models jointly. To alleviate this, a thin adapter is introduced that enables the recomposition of pre-trained sub-models without retraining them. In this way, a suite of new objection detection models are built with distinct accuracy and compute profiles at minimal training cost. These new models outperform by up to 3% in mAP on the COCO dataset at up to 99% less training cost when compared to conventional training approaches.