Image upsampling using one or more neural networks

Apparatuses, systems, and techniques are presented to reconstruct one or more images. In at least one embodiment, one or more neural networks are used to upsample one or more images based, at least in part, on one or more brightness values. The brightness values may include one or more exposure valu...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Loudon Cohen, Gregory Massal, Pekka Janis
Format: Patent
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Apparatuses, systems, and techniques are presented to reconstruct one or more images. In at least one embodiment, one or more neural networks are used to upsample one or more images based, at least in part, on one or more brightness values. The brightness values may include one or more exposure values calculated for at least a current input image of the one or more images. The one or more exposure values may be used to reduce a colour range of the current input image and a prior upsampled image of the one or more images. The one or more neural networks may be used to infer blending weights for corresponding pixels of at least the current input image and the prior upsampled image and may also be used to increase a colour range of one or more output images generated based at least in part upon the blending weights for the current input image and the prior upsampled image. The blending weights may be applied to colour values from the current input image and the prior upsampled image and may be determined in part using an accumulation of values determined using a rendering application-provided exposure value.