Nerve radiation field-based unbounded scene real-time rendering method, system and equipment

The invention belongs to the field of artificial intelligence, computer vision and computer graphics, particularly relates to an unbounded scene real-time rendering method, system and equipment based on a neural radiation field, and aims to solve the problems that feature distribution of an existing...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: PEI YAO, CUI MING, HE BIN, XU JIE, QIAO LEI, WANG YUE, WEI JIANWEI, SUN SIYUAN, ZHANG BAIPENG, ZHANG ZAOSHI, LYU YANNA, YANG FENGFENG, FENG XING, YAN RUYU, GAO PENGRUI, YANG ZIJIAN, ZHANG ZHIHENG, WANG CHAOPENG, FANG SHUN, ZHOU HAO
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 belongs to the field of artificial intelligence, computer vision and computer graphics, particularly relates to an unbounded scene real-time rendering method, system and equipment based on a neural radiation field, and aims to solve the problems that feature distribution of an existing NeRF method is wasted, the model training time is long, only bounded scene rendering can be processed, and control flow divergence exists. The method comprises the following steps: acquiring any image with known pose in a boundless scene as an input image; the method comprises the following steps: standardizing points in an input image, and after standardization, carrying out position coding in combination with a pre-constructed Hash index table; and inputting the feature vectors after position coding into a pre-constructed attribute prediction network to obtain the volume density and RGB color of points in the input image, and further obtaining an image after new view rendering. According to the method, the memor