3D Concept Grounding on Neural Fields

In order to answer questions about images, current methods rely on 2D segmentation masks obtained from supervised models.

from Technology Org https://ift.tt/BLSIWOf
via IFTTT