notesum.ai

Published at November 27

Helvipad: A Real-World Dataset for Omnidirectional Stereo Depth Estimation

cs.CV

cs.AI

cs.RO

Released Date: November 27, 2024

Authors: Mehdi Zayene¹, Jannik Endres², Albias Havolli¹, Charles Corbière, Salim Cherkaoui¹, Alexandre Kontouli¹, Alexandre Alahi¹

Aff.: ¹Ecole Polytechnique Fédérale de Lausanne, Switzerland; ²Ecole Polytechnique Fédérale de Lausanne, Switzerland; TU Darmstadt, Germany

Arxiv: http://arxiv.org/abs/2411.18335v1

Refer to caption

Method	Type	Disparity (°)			Depth (m)			Runtime (s)
		MAE $\downarrow$	RMSE $\downarrow$	MARE $\downarrow$	MAE $\downarrow$	RMSE $\downarrow$	MARE $\downarrow$
PSMNet [6]	stereo	0.33	0.54	0.20	2.79	6.17	0.29	0.66
360SD-Net [28]	360° stereo	0.21	0.42	0.18	2.14	5.12	0.15	0.63
IGEV-Stereo [32]	stereo	0.22	0.41	0.17	1.85	4.44	0.15	$\bm{0.24}$
360-IGEV-Stereo	360° stereo	$\bm{0.18}$	$\bm{0.39}$	$\bm{0.15}$	$\bm{1.77}$	$\bm{4.36}$	$\bm{0.14}$	0.25