Revisiting Point Cloud Completion: Are We Ready For The Real-World?

TL;DR

This work tackles point cloud completion in real-world, noisy, multi-sensor settings where synthetic benchmarks fall short. It reveals that real-world clouds harbor richer topological structure, detectable via Persistent Homology, and introduces RealPC, a large paired dataset capturing industrial scenes with ground-truth completions. The authors propose TopODGNet, a topology-regularized completion framework leveraging 0-dim PH priors, and BOSHNet, a computationally efficient Homology Sampler that provides backbone guidance without heavy PH computations. Experiments show existing methods underperform on RealPC, while PH-informed priors improve topological fidelity and completion quality, underscoring the need to rethink PC completion for real-world data and paving the way for topology-guided industrial 3D perception. The work also demonstrates kit-level applicability by extending 0-dim PH priors to KITTI scenes, highlighting broader practical impact for autonomous systems and robotics.

Abstract

Point clouds acquired in constrained, challenging, uncontrolled, and multi-sensor real-world settings are noisy, incomplete, and non-uniformly sparse. This presents acute challenges for the vital task of point cloud completion. Using tools from Algebraic Topology and Persistent Homology (PH), we demonstrate that current benchmark object point clouds lack rich topological features that are integral part of point clouds captured in realistic environments. To facilitate research in this direction, we contribute the first real-world industrial dataset for point cloud completion, RealPC - a diverse, rich and varied set of point clouds. It consists of ~ 40,000 pairs across 21 categories of industrial structures in railway establishments. Benchmark results on several strong baselines reveal that existing methods fail in real-world scenarios. We discover a striking observation - unlike current datasets, RealPC consists of multiple 0- and 1-dimensional PH-based topological features. We prove that integrating these topological priors into existing works helps improve completion. We present how 0-dimensional PH priors extract the global topology of a complete shape in the form of a 3D skeleton and assist a model in generating topologically consistent complete shapes. Since computing Homology is expensive, we present a simple, yet effective Homology Sampler guided network, BOSHNet that bypasses the Homology computation by sampling proxy backbones akin to 0-dim PH. These backbones provide similar benefits of 0-dim PH right from the start of the training, unlike similar methods where accurate backbones are obtained only during later phases of the training.

Figures (12)

• Figure 1: (a) to (c) Progression of filtration on a PC over different spatial resolutions as the distance threshold increases moor2020topological. (d) Birth and death of $k$-dim topological features documented in the form of a persistence diagram, i.e., $(b_i,d_i)$ pairs, so that each point corresponds to a homology which is born at $b_i$ and dies at $d_i$.
• Figure 2: Scene-level PCs from different acquisition techniques (a) sncfNuagePoints, (b) cserep2022hungarian, (c) qiu2024whu, and (d) ton2022labelled.
• Figure 3: Object-level training dataset creation methodology. Input to (A) shows the segmented industrial structures from scene-level PCs. Figure (B) shows how manual inspection was done to extract ground truth data. Output of (C), (D), and (E) are the three variants of sparse and incomplete PCs.
• Figure 4: Sub-optimal completion results on a RealPC PC using the best-performing completion baseline SnowflakeNet (Table \ref{['tab:baseline_comparison_21']}).
• Figure 5: TopODGNet. We calculate 0-dim$\mathcal{PH}$ based topological priors over sparse seeds and integrate it into the loss function. It enables completion along a topologically consistent skeleton.