PatchPoison: Poisoning Multi-View Datasets to Degrade 3D Reconstruction

TL;DR

PatchPoison is a simple yet effective dataset poisoning technique that inserts small adversarial patches into images to disrupt 3D reconstruction pipelines without affecting human perception.

Contribution

It introduces a lightweight, practical method for preventing unauthorized 3D scene reconstruction by corrupting feature matching with minimal visual disturbance.

Findings

Increases reconstruction error by 6.8x on NeRF-Synthetic benchmark.

Uses a 12x12 pixel patch to effectively disrupt SfM pipelines.

Does not require modifications to existing 3D reconstruction pipelines.

Abstract

3D Gaussian Splatting (3DGS) has recently enabled highly photorealistic 3D reconstruction from casually captured multi-view images. However, this accessibility raises a privacy concern: publicly available images or videos can be exploited to reconstruct detailed 3D models of scenes or objects without the owner's consent. We present PatchPoison, a lightweight dataset-poisoning method that prevents unauthorized 3D reconstruction. Unlike global perturbations, PatchPoison injects a small high-frequency adversarial patch, a structured checkerboard, into the periphery of each image in a multi-view dataset. The patch is designed to corrupt the feature-matching stage of Structure-from-Motion (SfM) pipelines such as COLMAP by introducing spurious correspondences that systematically misalign estimated camera poses. Consequently, downstream 3DGS optimization diverges from the correct scene geometry. On the NeRF-Synthetic benchmark, inserting a 12 X 12 pixel patch increases reconstruction error by 6.8x in LPIPS, while the poisoned images remain unobtrusive to human viewers. PatchPoison requires no pipeline modifications, offering a practical, "drop-in" preprocessing step for content creators to protect their multi-view data.