Reproduce “First Competition on Presentation Attack Detection on ID Card” 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
Juan E. Tapia, Naser Damer, Christoph Busch, Juan M. Espin, Javier Barrachina, Alvaro S. Rocamora, Kristof Ocvirk, Leon Alessio and 7 more
The inaugural PAD-IDCard competition at IJCB 2024 established a benchmark for Presentation Attack Detection (PAD) on ID cards, evaluating algorithms in a cross-dataset scenario. The organizers' private-data-trained baseline achieved an AV_rank of 9.10%, substantially outperforming the best competitor's AV_rank of 74.30%, highlighting the critical role of data access and diversity.
Key insights
- Access to extensive and diverse training data is paramount for high performance and generalization in ID card PAD, as evidenced by the superior performance of organizers' private-data-trained baselines.
- Screen attacks were identified as the most challenging Presentation Attack Instrument Species (PAIS) to detect, often attributed to the high resolution and variability of modern digital displays.
- Generalization capabilities are significantly hampered by the variability in ID card designs across countries, with non-ICAO compliant cards proving substantially more difficult for models to process.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.