The reviewed record of science sign in
Pith

arxiv: 2604.07010 · v1 · submitted 2026-04-08 · cs.CV

Synthetic Dataset Generation for Partially Observed Indoor Objects

Jelle Vermandere , Maarten Bassier , Maarten Vergauwen This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-05-10 19:09 UTCgrok-4.3open to challenge →

classification cs.CV
keywords synthetic dataset3D scanningindoor scenespoint cloudsobject completionscene reconstructionvirtual simulation
0
0 comments X

The pith

A Unity-based virtual scanning framework generates synthetic 3D indoor datasets containing partial point clouds and complete ground truth.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Acquiring real-world 3D scan datasets with accurate ground truth for occluded areas is costly and time-consuming. The paper presents a virtual scanning framework in Unity that simulates real scanner behavior through ray-based scanning from virtual viewpoints to model visibility, occlusion, and noise effects. Panoramic images assign colors to the generated point clouds. This framework is paired with procedural generation of diverse indoor scenes to enable scalable dataset creation. It yields the V-Scan dataset of synthetic indoor scans, object-level partial point clouds, voxel-based occlusion grids, and full ground-truth geometry for training reconstruction and completion models.

Core claim

The authors present a Unity-implemented virtual scanning framework that simulates real-world scanners via ray-based scanning from virtual viewpoints to model sensor visibility and occlusion effects, along with distance-dependent noise and panoramic color assignment. Integrated with a procedural indoor scene generation pipeline, the system produces the V-Scan dataset consisting of synthetic indoor scans together with object-level partial point clouds, voxel-based occlusion grids, and complete ground-truth geometry to support learning-based methods for scene reconstruction and object completion.

What carries the argument

Ray-based scanning from virtual viewpoints integrated with procedural indoor scene generation, which enables realistic modeling of sensor visibility and occlusion without direct mesh sampling.

If this is right

  • Allows creation of large-scale datasets without the expense of real scanning hardware.
  • Provides ground truth for occluded regions that real scans cannot easily capture.
  • Supports training on diverse room layouts and furniture arrangements generated procedurally.
  • Offers configurable parameters to tune the simulation to different scanner types.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Models trained on this data could accelerate development of 3D reconstruction tools for applications like virtual reality or autonomous navigation.
  • The method might extend to generating datasets for other sensor types or non-indoor environments.
  • Validation on real data transfer would be needed to confirm the simulation's fidelity for practical use.

Load-bearing premise

The ray-based scanning simulation with its configurable parameters and noise models is sufficiently similar to real-world scanners to generate useful training data for learning-based methods.

What would settle it

A direct comparison where a reconstruction model is trained on V-Scan and evaluated on real indoor scans; poor transfer performance would indicate the synthetic data does not adequately replicate real scanner characteristics.

Figures

Figures reproduced from arXiv: 2604.07010 by Jelle Vermandere, Maarten Bassier, Maarten Vergauwen.

Figure 1
Figure 1. Figure 1: Virtual scanner overview with: input [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Virtual scanner setup in Unity with configurable [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The scan vectors that are generated at the start of the [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: A equirectangular projection of the panoramic image [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 7
Figure 7. Figure 7: An overview of the different layouts, with top-down and perspective views respectively [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
read the original abstract

Learning-based methods for 3D scene reconstruction and object completion require large datasets containing partial scans paired with complete ground-truth geometry. However, acquiring such datasets using real-world scanning systems is costly and time-consuming, particularly when accurate ground truth for occluded regions is required. In this work, we present a virtual scanning framework implemented in Unity for generating realistic synthetic 3D scan datasets. The proposed system simulates the behaviour of real-world scanners using configurable parameters such as scan resolution, measurement range, and distance-dependent noise. Instead of directly sampling mesh surfaces, the framework performs ray-based scanning from virtual viewpoints, enabling realistic modelling of sensor visibility and occlusion effects. In addition, panoramic images captured at the scanner location are used to assign colours to the resulting point clouds. To support scalable dataset creation, the scanner is integrated with a procedural indoor scene generation pipeline that automatically produces diverse room layouts and furniture arrangements. Using this system, we introduce the \textit{V-Scan} dataset, which contains synthetic indoor scans together with object-level partial point clouds, voxel-based occlusion grids, and complete ground-truth geometry. The resulting dataset provides valuable supervision for training and evaluating learning-based methods for scene reconstruction and object completion.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper presents a Unity-based virtual scanning framework that simulates real-world 3D scanners via ray casting with configurable parameters for resolution, range, and distance-dependent noise, augmented by panoramic color assignment from virtual viewpoints. Integrated with a procedural pipeline for generating diverse indoor scenes, the system produces the V-Scan dataset containing synthetic scans, object-level partial point clouds, voxel occlusion grids, and complete ground-truth geometry to provide supervision for learning-based 3D scene reconstruction and object completion.

Significance. If the simulated data can be shown to match real scanner characteristics, the framework would offer a scalable, low-cost method for generating large paired partial-complete datasets with known ground truth for occluded regions, addressing a key bottleneck in training models for indoor 3D vision tasks and enabling more reproducible experiments.

major comments (2)
  1. [Abstract] Abstract: the central claim that the framework generates 'realistic' synthetic datasets via ray-based scanning and distance-dependent noise is asserted without any quantitative validation, such as distribution matching on point density, occlusion statistics, noise histograms, or visibility patterns against real scanner outputs from ScanNet or Matterport3D.
  2. [Results/Evaluation] The manuscript describes the simulation parameters and panoramic coloring but provides no empirical results section demonstrating that V-Scan improves downstream task performance (e.g., reconstruction accuracy) when used for training compared to real data or other synthetic baselines.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the two major comments point by point below, focusing on how the work can be strengthened without overstating its current scope.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the framework generates 'realistic' synthetic datasets via ray-based scanning and distance-dependent noise is asserted without any quantitative validation, such as distribution matching on point density, occlusion statistics, noise histograms, or visibility patterns against real scanner outputs from ScanNet or Matterport3D.

    Authors: We agree that the abstract employs the term 'realistic' to characterize the output of the ray-based simulation with configurable parameters drawn from real scanner specifications. The manuscript explains the simulation mechanics (ray casting for visibility/occlusion and distance-dependent noise) but does not perform quantitative distribution matching or statistical comparisons against real datasets such as ScanNet or Matterport3D. This is a fair observation. In the revised manuscript we will replace 'realistic' with 'physically motivated' in the abstract and add a short limitations paragraph noting the absence of direct empirical validation against real scanner statistics, while highlighting that the framework's parameters are user-configurable to approximate specific devices. revision: partial

  2. Referee: [Results/Evaluation] The manuscript describes the simulation parameters and panoramic coloring but provides no empirical results section demonstrating that V-Scan improves downstream task performance (e.g., reconstruction accuracy) when used for training compared to real data or other synthetic baselines.

    Authors: The manuscript's primary contribution is the Unity-based virtual scanning framework together with the released V-Scan dataset of paired partial scans and complete geometry. It does not contain an empirical evaluation section or claim specific performance gains on downstream tasks such as scene reconstruction or object completion. We view the dataset as a resource that enables such experiments rather than a demonstration of them. To respond to this comment we will add a brief 'Usage and Potential Applications' subsection that includes a minimal illustrative experiment (e.g., training a simple completion network on V-Scan and reporting basic metrics), together with guidance on how the data can be used for comparative studies, while making clear that a full benchmark against real data remains future work. revision: partial

Circularity Check

0 steps flagged

No circularity: paper presents implementation and dataset without derivations or predictions

full rationale

The manuscript describes a Unity-based virtual scanning framework and the resulting V-Scan dataset. No equations, fitted parameters, predictions, or derivation steps appear in the provided text. The contribution is the system itself (ray-casting simulation plus procedural scene generation), not a claim that some output is derived from prior results. No self-citations, ansatzes, or uniqueness theorems are invoked in any load-bearing way. The reader's circularity score of 0.0 is therefore confirmed; the work is self-contained as an engineering contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work relies on standard assumptions in simulation and procedural modeling without introducing new physical entities or fitted parameters beyond user-configurable scanner settings.

axioms (2)
  • domain assumption Ray casting from virtual viewpoints accurately simulates real scanner visibility and occlusion effects
    Invoked to justify the ray-based scanning approach for modeling sensor behavior.
  • domain assumption Procedural generation produces sufficiently diverse and realistic indoor room layouts and furniture arrangements
    Used to support scalable dataset creation.

pith-pipeline@v0.9.0 · 5508 in / 1303 out tokens · 77220 ms · 2026-05-10T19:09:22.052788+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

  1. [1]

    McHugh, and Vincent Vanhoucke

    https://isprs-archives.copernicus.org/articles/XLVI-5-W1- 2022/59/2022/. Downs, L., Francis, A., Koenig, N., Kinman, B., Hickman, R., Reymann, K., McHugh, T. B., Vanhoucke, V ., 2022. Google Scanned Objects: A High-Quality Dataset of 3D Scanned Household Items. arXiv:2204.11918 [cs]. Leica BLK360|3D Laser Scanner, n.d. Leica ScanStation P40 / P30 - High-D...

    work page arXiv 2022

  2. [2]

    2024.doi:10.48550/arXiv.2303.14207

    DiffuScene: Denoising Diffusion Models for Generative Indoor Scene Synthesis. arXiv:2303.14207 [cs]. Vermandere, J., Bassier, M., Vergauwen, M., 2025. Geo- metry and Texture Completion of Partially Scanned 3D Objects Through Material Segmentation:.Proceedings of the 20th In- ternational Joint Conference on Computer Vision, Imaging and Computer Graphics Th...

    work page arXiv 2025