Verfahren und Vorrichtung zur Segmentierung Zeitlicher Aktionen, Modelltrainingsverfahren und -Vorrichtung und Speichermedium

Details

Veröffentlichungs-Nr.

EP4773102

Anmeldetag

10. Januar 2025

Veröffentlichung

24. Juli 2025

Rechtsraum

EP

IPC

G06V20/40

Abstract

This disclosure provides a temporal action segmentation method and apparatus, a model training method and apparatus, as well as a storage medium, which relates to the technical field of artificial intelligence. The method for training a temporal action segmentation model of the present disclosure includes: performing high-dimensional feature extraction on data of each modality in sample data to obtain a first feature for the each modality; obtaining an intermediate loss function based on the first feature through dimensionality reduction, and obtaining second feature for the each modality through dimensionality restoration; obtaining a motion representation based on the second feature and a supplementary feature, by fusing features of different modalities; and processing the motion representation through a category encoder network and a boundary encoder network separately, to obtain an action category prediction vector and a temporal boundary prediction vector, and determining an action category prediction result and a prediction loss function, wherein parameters of the temporal action segmentation model are adjusted based on the intermediate loss function and the predicted loss function until training is completed.

Anmelder

Firmen

Beijing Wodong Tianjun Information Technology Co., Ltd.
Jingdong Technology Holding Co., Ltd.

Land

🇨🇳 China

Fachgebiete