Core Concepts
TPVD is a novel framework that explicitly models 3D geometry by decomposing point clouds into three 2D views, enhancing depth completion accuracy.
Abstract
Depth completion is crucial for autonomous driving, reconstructing precise 3D scenes from sparse depth data.
TPVD decomposes point clouds into three 2D views, improving geometric understanding.
TPV Fusion and GSPN enhance geometric consistency and refinement.
TPVD outperforms existing methods on various datasets like KITTI, NYUv2, and SUN RGBD.
TOFDC dataset collected using TOF sensors and smartphones enhances depth completion research.
Stats
Depth completion is vital for autonomous driving as it involves reconstructing the precise 3D geometry of a scene from sparse and noisy depth measurements.
Most previous methods focus on 2D feature space to learn depth representations, leading to a lack of 3D geometric information.
TPVD outperforms existing methods on KITTI, NYUv2, and SUN RGBD datasets.
TOFDC dataset is acquired by time-of-flight (TOF) sensor and color camera on smartphones.