Reproduce “DINO-VPT: Hierarchical Visual Prompt Tuning for Joint Physical-Digital Face Anti-Spoofing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pierre Gallin-Martel, Mika Feng, Koichi Ito, Takafumi Aoki
DINO-VPT is a vision-only framework that leverages hierarchical visual prompt tuning and a dynamic routing network to adapt a DINOv2 backbone for Unified Physical-Digital Face Anti-Spoofing. This approach achieves an ACER of 0.07% on intra-domain data and 0.63% on cross-domain data, outperforming prior multimodal methods with improved computational efficiency.
Key insights
- Dynamically routed hierarchical visual prompt tuning enables a vision-only foundation model to effectively disentangle diverse physical and digital spoofing cues, bypassing multimodal fusion.
- Input-conditioned prompt injection through a Prompt Routing Network is critical for achieving robust cross-modality generalization by adaptively activating relevant prompt subsets.
- Dynamically simulating spoofing cues and relabeling them during training, guided by a hierarchical taxonomy, significantly boosts robustness and balances detection performance across attack types.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.