Reproduce “GBU-Palm: A Multimodal Video Dataset and Benchmark for Palm Presentation Attack Detection” 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
Yingjie Ma, Zitong Yu, Wei Jia, Ajay Kumar, Linlin Shen
GBU-Palm introduces a large-scale multimodal video dataset and benchmark for palm presentation attack detection, designed to overcome the limitations of static images and single-spectrum sensing in prior datasets. It facilitates systematic evaluation of video-based PAD methods' robustness across various environmental conditions, attack types, and multimodal inputs.
Key insights
- The robustness of palm PAD models to cross-environment generalization is highly architecture-dependent, with varying performance degradation across different video backbones.
- Multimodal information fusion (e.g., RGB-NIR) does not universally improve PAD performance; its benefits are specific to the architecture and task, with some models showing increased error rates.
- The reliance on explicit temporal ordering of frames varies substantially across different video architectures, with some models demonstrating limited sensitivity to shuffled or reversed frame sequences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.