notesum.ai

Published at November 26

DepthCues: Evaluating Monocular Depth Perception in Large Vision Models

cs.CV

Released Date: November 26, 2024

Authors: Duolikun Danier¹, Mehmet Aygün, Changjian Li¹, Hakan Bilen¹, Oisin Mac Aodha¹

Aff.: ¹University of Edinburgh

Arxiv: http://arxiv.org/abs/2411.17385v1

Refer to caption

Model	NYUv2 Acc. (%) $\uparrow$	DIW WHDR (%) $\downarrow$
DINOv2	87.78	11.99
DINOv2+DC	87.06	11.95
concat(DINOv2, noise)	87.56	12.20
concat(DINOv2, DINOv2+DC )	88.46	11.72
CLIP	43.78	35.25
CLIP+DC	43.59	35.45
concat(CLIP, noise)	43.38	35.39
concat(CLIP, CLIP+DC )	44.32	33.53