RAW: Robust Avatar Watermarking -- Benchmarking and Baseline

2026-05-17

AI Summary

This study addresses the insufficient robustness of existing digital avatar watermarking methods under common post-processing operations such as background replacement and cropping, as well as the absence of standardized evaluation benchmarks tailored to avatar-specific attacks. To bridge this gap, the authors introduce RAW, the first public benchmark comprising 50 synthesized videos and six categories of realistic attacks, and propose WALT, a novel watermarking approach that embeds watermarks into the UV texture space via 3D facial reconstruction and leverages deep learning for robust encoding and decoding. Experimental results demonstrate that WALT achieves watermark recovery rates of 95.6% and 92.4% under background removal and scaling attacks, respectively, substantially outperforming state-of-the-art methods. This work establishes a new paradigm and provides a standardized platform for future research in avatar watermarking.

Abstract

Digital avatar watermarking presents unique challenges: avatars are routinely post-processed with background replacement, reframing, and format conversion before deployment. We introduce \textbf{RAW} (Robust Avatar Watermarking), a benchmark comprising 50 synthetic avatar videos from 5 commercial providers and 6 attacks simulating real-world avatar workflows. Evaluating 7 existing methods reveals that avatar-specific attacks such as background removal significantly degrade watermark recovery. We propose \textbf{WALT} (Watermarking Avatars with Learned Textures), which embeds watermarks in UV texture space via 3D face reconstruction. WALT achieves the highest robustness to zoom attacks (92.4\%) while maintaining strong performance on background removal (95.6\%). We release our benchmark to facilitate research into avatar-specific watermarking.

Problem

• avatar watermarking
• robustness
• background removal
• post-processing
• digital avatar

Innovation

• avatar watermarking
• UV texture embedding
• 3D face reconstruction
• robustness benchmark
• WALT