PaperlayerView exact revision on arXiv

Source-linked AI summary

VR-Themis: A Scalable Framework for Virtual Reality Application Clone Detection

Gengyang Xu, Hanyang Guo, Hong-Ning Dai, Weizhi Meng

arXiv:2608.13290v1cs.CRcs.SE

TL;DR

VR app cloning threatens developers and users, while mobile-oriented detectors miss VR-specific scene and asset characteristics. VR-Themis addresses this gap with coarse-grained statistical clustering followed by fine-grained HOB similarity analysis, detecting 307 clone apps among 4,277 VR apps without false positives. The study is evaluated on Unity-based Meta Quest apps, and its stated scope excludes other engines and some dynamically loaded resources.

  • Problem

    Existing mobile clone detectors fail to capture the essential 3D scene, object, and Behaviour features of VR apps, leaving proprietary VR clone detection insufficiently addressed.

  • Method

    VR-Themis uses coarse-grained clustering of retrievable statistical features followed by fine-grained comparison with HOB metrics covering hierarchy, objects, and script Behaviours.

  • Results

    307 clone apps were detected from 4,277 VR apps without false positives.

  • Takeaways & Limitations

    VR-Themis provides a VR-specific clone detection framework that the experiments report as effective and scalable for the collected dataset.

  • Takeaways & Limitations

    The study focuses on Unity-based Meta Quest apps and excludes VR apps that dynamically load resources at runtime, while future work targets other engines and platforms.

Abstract

from arXiv · show

Repackaging of mobile applications (aka app cloning) not only threatens the security and privacy of mobile users but also infringes upon the copyright of the original app developers. However, existing detection methods that primarily focus on mobile platforms (such as Android) fail to capture the essential features of virtual reality (VR). Consequently, they are inadequate for effectively detecting cloned VR apps, which have often been targeted by illegal users in the VR market. Considering the unique features of VR apps, this paper proposes a two-stage app clone detection framework, namely VR-Themis, based on \emph{Hierarchy-Object-Behaviour} (HOB). Firstly, VR-Themis exploits the coarse-grained stage to cluster apps based on their retrievable statistical features, making this tool scalable to large-scale VR app datasets. Then, in the fine-grained stage, VR-Themis performs in-depth analysis of the suspicious apps (identified in the first stage) by calculating similarity using our defined \emph{HOB metrics}. Our extensive experiments indicate that VR-Themis successfully detects 307 suspected clone apps from the collected 4,277 VR apps without false positives, demonstrating its effectiveness and scalability.

1 Introduction

VR-Themis addresses the mismatch between mobile-oriented clone detectors and VR’s 3D scene, object, and Behaviour features with a scalable two-stage framework. It combines coarse-grained clustering with fine-grained HOB comparisons and evaluates the approach on a large VR app dataset.

  • Motivation: VR app development emphasizes immersive 3D scenes and assets, which conventional mobile clone detectors do not adequately capture.Mobile detectors typically rely on code, UI layouts, and resource metadata, while VR apps include scene hierarchies, visual elements, and interactions.
  • Approach: VR-Themis first groups apps using retrievable statistical features to reduce the complexity of pairwise comparisons.The coarse-grained stage makes large-scale analysis more scalable by narrowing the apps requiring detailed comparison.
  • Approach: The HOB model and metrics compare VR-specific hierarchy, object-level, and script-driven Behaviour characteristics.HOB metrics integrate Hierarchy Edit Distance, GameObject Node Distance, and Script Behaviour Similarity for fine-grained analysis.
  • Contributions: The study presents an automatic detector and identifies proprietary VR app clone detection as a previously unstudied problem.The authors describe VR-Themis as the first study on proprietary VR app clone detection.
  • Evaluation: 4,277 Unity-based VR apps were used in extensive experiments, which reported effective and efficient clone detection with high accuracy.The dataset was collected from diverse sources, including side-loading channels not considered in previous research.

2 Background and Threat Model

VR app cloning threatens developers and users in an ecosystem built around immersive 3D content rather than conventional mobile interfaces. The threat model defines clones by unauthorized ownership and assumes largely static APK-contained resources while allowing adversarial modifications and unofficial distribution.

  • Background: VR cloning can undermine developers, increase app-market maintenance burdens, and threaten user privacy and security.Pirated repackaged apps continue to circulate through communities that distribute unauthorized downloads and installation methods.
  • VR Application Development: VR development differs from mobile development by emphasizing high-quality 3D assets and immersive 3D scenes.VR operating systems may build on Android, but VR applications have distinct development requirements centered on immersion.
  • VR Application Development: A VR scene is represented as a hierarchy tree whose nodes are GameObjects linked by parent-child relationships.GameObjects contain components that provide the properties needed for visual, functional, and interactive behavior.
  • Threat Model: VR-Themis defines a clone as substantially similar apps in architecture and fundamental assets distributed under distinct ownership without authorization.This definition distinguishes clone detection from code reuse detection, which does not emphasize scene hierarchies and digital assets.
  • Threat Model: The adversary may decompile APKs, modify scripts, 3D models, and scene hierarchies, then distribute clones through unofficial channels or masquerade as developers.The assumed capabilities include extracting scene hierarchies, assets, and C# scripts before modification and redistribution.
  • Threat Model: The framework assumes most scene hierarchies and assets are stored statically in APKs, excluding apps that load resources dynamically at runtime.It also assumes adversaries can make minor modifications to cloned apps to circumvent detection.

3 Hierarchy-Object-Behaviour Similarity Metric for VR App Clone Detection

The HOB framework represents VR apps through scene hierarchies, GameObjects, and script-driven Behaviours, then compares clones using complementary structural, object-level, and Behaviour-level metrics.

  • 3.1 Hierarchy-Object-Behaviour Model: The HOB model encodes VR apps as scene hierarchies of GameObjects containing components and script-driven Behaviours.It captures VR-specific structure, visual and functional objects, and custom interactive functionality.
  • 3.2 HOB Metrics: Multi-Dimensional VR App Similarity Measures: Hierarchy Edit Distance measures the minimum-cost transformations needed to convert one VR scene hierarchy tree into another.Insertion and deletion each cost 1, while node modification uses GameObject component differences.
  • 3.2 HOB Metrics: Multi-Dimensional VR App Similarity Measures: Together, HED, GND, and SBS capture structural transformations, object-level variations, and Behaviour-level modifications in fine-grained VR app comparisons.The metric suite is designed to address VR characteristics not represented by mobile-oriented clone features.
  • 3.2 HOB Metrics: Multi-Dimensional VR App Similarity Measures: GameObject Node Distance quantifies object-level differences using component-set Jaccard distance, with different 3DObject meshes assigned distance 1.The mesh exception reflects the substantially different visual information represented by different meshes.
  • 3.2 HOB Metrics: Multi-Dimensional VR App Similarity Measures: Script Behaviour Similarity compares script-driven Behaviours using feature-vector cosine similarity, treating components with similarity at least 0.8 as identical.The threshold follows a common mobile repackaging-detection indicator based on sharing 80% of code.

4 Design of VR-Themis

VR-Themis targets Unity-based Meta Quest VR apps through a two-stage workflow: scalable coarse-grained clustering followed by fine-grained HOB-based similarity analysis. Its design captures scene hierarchy, 3D-object, and script-Behaviour characteristics using specialized extraction and comparison procedures.

  • 4.1 Scope: VR-Themis focuses primarily on Unity-developed VR apps for Meta Quest series headsets.The scope reflects Meta Quest’s market dominance and Unity’s prevalence among top-selling Quest games.
  • 4.2 Coarse-grained Processing: The coarse-grained stage clusters apps using ten statistical features spanning hierarchy, object-level, and Behavioural aspects.The features include scene levels, GameObjects, AnimationClips, Animators, Materials, Sprites, Texture2Ds, AudioClips, meshes, and MonoBehaviours.
  • 4.3 Fine-grained Processing: Fine-grained processing compares suspicious apps through extracted scene hierarchies and VR assets using structural, object-level, and Behavioural HOB metrics.The workflow organizes a scene hierarchy tree and integrates HED, GND, and SBS for multidimensional similarity assessment.
  • 4.3 Fine-grained Processing: AssetStudio.Fine extracts scene hierarchies, GameObject types, components, meshes, and scripts for the fine-grained comparison steps.Unity YAML Class ID mappings help identify built-in component types during asset parsing.
  • 4.3 Fine-grained Processing: Mesh processing applies mesh decimation to reduce vertices and faces while minimizing shape changes, improving efficiency over direct complex-mesh comparison.The method addresses the processing cost of hash-based mesh comparison for large experimental datasets.
  • 4.3 Fine-grained Processing: Script processing uses SBS cosine similarity over script metadata to compare Behavioural similarity across Mono-based and IL2CPP Unity backends.The feature vectors use method declarations and data attributes, supporting similarity detection after lightweight modifications.

5 Implementation and Evaluation

VR-Themis was evaluated on a 4,277-app Unity-based Meta Quest dataset using coarse-grained clustering, fine-grained similarity analysis, and manual clone verification. The evaluation shows substantial comparison reduction, faster mesh processing, and detection of suspected clones without false positives.

  • 5.1 Dataset: 4,277 Unity-based Meta Quest APKs were collected from five sources, with sizes ranging from 18.26 MB to 2,305.95 MB and most below 400 MB.The dataset was described as the largest VR APK dataset constructed for clone detection at the time.
  • 5.2 Coarse-grained Detection: DBSCAN clustering reduced 9,144,226 possible app pairs to 416,385 suspected clone pairs, approximately 4.55% of the original comparisons.Feature extraction took 5,810.44 seconds, clustering took 0.1024 seconds, and produced 198 clusters.
  • 5.2 Coarse-grained Detection: A 100-app validation subset achieved 100% recall and 80.65% precision with ϵ = 0.1 for DBSCAN.The parameter was selected after manually installing and executing sampled applications to establish clone truths.
  • 5.3 Fine-grained Detection: 53.54% average mesh-processing time reduction was obtained through mesh decimation on 8,000 meshes from 500 randomly selected APKs.Decimation ratios were 0.4 for medium-size meshes and 0.2 for large-size meshes.
  • 5.3 Fine-grained Detection: Mesh decimation preserved identical meshes and produced no false positives in an evaluation of 3,000 mesh files.The fine-grained similarity threshold was set to 80% based on false-positive and false-negative rate analysis.
  • 5.4 Cloning Verification and Accuracy Analysis: Manual review of all identified clones found zero false positives, while additional theme-based review found no apparent undetected clones.Verification included signature filtering followed by installation, execution, and assessment of visual design and game mechanics.

6 Discussion

The discussion examines shared assets and engine or device scope boundaries. VR-Themis combines structural and behavioural similarity with visual similarity to reduce mistaken clone flags, but some reuse cases remain difficult to distinguish.

  • Handling Shared Assets and Templates: Structural HED and behavioural SBS similarities supplement visual similarity when apps share common 3D models, animations, or interaction templates.The authors state that this mechanism prevents ordinary shared assets and templates from being mistakenly flagged as clones.
  • Handling Shared Assets and Templates: Lightweight VR apps relying heavily on public assets or templates can blur the boundary between legitimate reuse and cloning.The authors propose comparing app assets with public libraries such as the Unity Asset Store in future work.
  • Scope Boundary: The current scope is limited to Unity-based VR apps and does not cover engines such as Unreal or libGDX or devices such as HTC VIVE Pro 2, PlayStation VR 2, and Pico 4.The authors plan to extend clone detection to additional platforms and engines.

7 Related Work

Related work covers mobile app clone detection and Unity-based application analysis. Prior mobile detectors compare code, resources, layouts, or combinations of these features, while Unity studies address reverse engineering and in-app purchasing vulnerabilities.

  • Mobile App Clone Detection: Android clone-detection methods use code dependence graphs, fuzzy hashing, resources, UI layouts, or combined resource-and-layout comparisons.Examples include DNADroid, DroidMOSS, FSquaDRA, DroidEagle, ResDroid, ViewDroid, and PiggyApp.
  • Unity-based Application: Unity-related studies have used static and dynamic analysis for malicious-app detection and static analysis for identifying in-app purchasing vulnerabilities.These studies concern Unity applications but address security analysis rather than proprietary VR clone detection.

8 Conclusion

The paper concludes that VR-Themis addresses VR-specific clone-detection challenges through two-stage HOB-driven analysis. On 4,277 VR apps, it detected 307 clone apps with no false positives.

  • Conclusion: VR-Themis targets VR app cloning, which threatens developer interests and user security while exposing gaps in mobile-focused detectors.Its framework combines coarse-grained statistical clustering with fine-grained HOB metric comparisons.
  • Conclusion: 307 clone apps were detected from 4,277 VR apps with no false positives.The conclusion presents this outcome as evidence of the framework’s effectiveness and scalability.
Loading 2608.13290v1…