GS-Checker: Tampering Localization for 3D Gaussian Splatting

TL;DR

GS-Checker is proposed, a novel method for locating tampered areas in 3DGS models that integrates a 3D tampering attribute into the 3D Gaussian parameters to indicate whether the Gaussian has been tampered.

Abstract

Recent advances in editing technologies for 3D Gaussian Splatting (3DGS) have made it simple to manipulate 3D scenes. However, these technologies raise concerns about potential malicious manipulation of 3D content. To avoid such malicious applications, localizing tampered regions becomes crucial. In this paper, we propose GS-Checker, a novel method for locating tampered areas in 3DGS models. Our approach integrates a 3D tampering attribute into the 3D Gaussian parameters to indicate whether the Gaussian has been tampered. Additionally, we design a 3D contrastive mechanism by comparing the similarity of key attributes between 3D Gaussians to seek tampering cues at 3D level. Furthermore, we introduce a cyclic optimization strategy to refine the 3D tampering attribute, enabling more accurate tampering localization. Notably, our approach does not require expensive 3D labels for supervision. Extensive experimental results demonstrate the effectiveness of our proposed method to locate the tampered 3DGS area.

Figures (7)

• Figure 1: Our proposed scenario for 3DGS tampering localization. When the owner creates a 3DGS model and releases it publicly over the web, malicious users can manipulate it using advanced 3D editing tools for illegal purposes. Upon the release of a tampered 3DGS, the public can use our proposed GS-Checker to identify the tampered areas, thereby protecting the integrity and ownership of the 3DGS model.
• Figure 2: Illustration of our proposed method. First, the 3D tampering attribute is integrated into the 3D Gaussian parameters and initialized via 3D voting. Next, a 3D contrastive mechanism is introduced to seek tampering cues by comparing the similarity of key attributes between 3D Gaussians. Finally, a cyclic optimization strategy is employed to iteratively refine the tampering attribute by projecting it back into the 2D space, enabling joint optimization across both 2D and 3D levels.
• Figure 3: Statistical distribution of tampered and authentic 3D Gaussian properties. Noticeable anomalies are observed in the tampered regions, where parameter distributions deviate from those of the authentic parts. These discrepancies serve as key indicators of tampering, motivating our approach to seek tampering traces at the 3D level.
• Figure 4: Qualitative results of 3DGS tampering localization in different scenes. Columns from left to right are: rendered images of tampered 3D scenes, ground-truth, SAFIRE kwon2025safire+SAGD sagd, SAFIRE kwon2025safire+SA3D sa3d and ours.
• Figure 5: Results on the authentic scene stump.