Target instance segmentation model establishment method based on automatic prompt learning and application thereof

The invention discloses a target instance segmentation model establishment method based on automatic prompt learning and application thereof, and belongs to the field of image segmentation, and the method comprises the steps: establishing a target segmentation network, and training the target segmen...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: NING KANGHUI, ZHAO ZHIJIAN, WANG GONGYAN, SHI XIANSHUO, GE SHICHEN, YANG WEIDONG, HUANG PENG
Format: Patent
Sprache:chi ; eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The invention discloses a target instance segmentation model establishment method based on automatic prompt learning and application thereof, and belongs to the field of image segmentation, and the method comprises the steps: establishing a target segmentation network, and training the target segmentation network through employing an image data set marked with a ground truth value label; the target segmentation network comprises: a feature embedding network comprising an image encoder in an SAM for obtaining image feature embedding of an input image; the contour mapping recognition network is used for embedding and predicting the category of each target, the position of a central point and the distance from the central point to n contour points according to image features; the automatic prompt learning network is used for screening out the most important part of contour points and mapping the most important part of contour points and the center point to a semantic space to obtain semantic embedding of point p