VVRec: Reconstruction Attacks on DL-based Volumetric Video Upstreaming via Latent Diffusion Model with Gamma Distribution

TL;DR

This work tackles privacy risks in DL-based volumetric video upstreaming by demonstrating a reconstruction attack that recovers original point clouds from intercepted intermediate results. It introduces VVRec, a four-module attacker leveraging latent diffusion models with a Gamma distribution (SLDM, LPGDM, LPG, RD) plus a Point Cloud Refinement (PCR) step to produce high-quality reconstructions. Across four volumetric datasets and multiple victim models, VVRec achieves substantial reconstruction quality, including color recovery, and outperforms prior baselines while exposing limited effectiveness of simple protective perturbations. The results highlight a concrete privacy threat in volumetric video streaming and motivate development of robust defense mechanisms.

Abstract

With the popularity of 3D volumetric video applications, such as Autonomous Driving, Virtual Reality, and Mixed Reality, current developers have turned to deep learning for compressing volumetric video frames, i.e., point clouds for video upstreaming. The latest deep learning-based solutions offer higher efficiency, lower distortion, and better hardware support compared to traditional ones like MPEG and JPEG. However, privacy threats arise, especially reconstruction attacks targeting to recover the original input point cloud from the intermediate results. In this paper, we design VVRec, to the best of our knowledge, which is the first targeting DL-based Volumetric Video Reconstruction attack scheme. VVRec demonstrates the ability to reconstruct high-quality point clouds from intercepted transmission intermediate results using four well-trained neural network modules we design. Leveraging the latest latent diffusion models with Gamma distribution and a refinement algorithm, VVRec excels in reconstruction quality, color recovery, and surpasses existing defenses. We evaluate VVRec using three volumetric video datasets. The results demonstrate that VVRec achieves 64.70dB reconstruction accuracy, with an impressive 46.39% reduction of distortion over baselines.

Figures (8)

• Figure 1: Video applications where the content provider is evolving, from filming companies to individual users, can be categorized into Downstream and Upstream Applications.
• Figure 2: Workflow of existing DL-based volumetric video streaming and privacy risks.
• Figure 3: Workflow of VVRec activates a reconstruction attack, including training NN models in VVRec and reconstructing the given intercepted intermediate results to a point cloud.
• Figure 4: Basic blocks of VVRec.
• Figure 5: Visualization of reconstruction result of VVRec and baselines on trace redandblack@8i during attacking victim encoder PCGCv1 on above Table 1.