Researchers have developed a new benchmark for long-range video presentation attack detection (PAD) systems, addressing the limitation of current systems that assume close-range subject capture. The benchmark includes over 374,000 frames from 261 identities, captured at distances from 1 to 100 meters under various conditions using both DSLR and mobile sensors. Four CNN architectures were fine-tuned and evaluated, revealing that while high-resolution DSLR cameras maintain performance, mobile sensors degrade significantly beyond 50 meters. EfficientNet achieved the highest accuracy at 98.48%, demonstrating robust generalization to unseen attack datasets. AI
IMPACT Establishes capture distance and sensor resolution as critical factors for AI face detection systems, potentially influencing future development and deployment strategies.
RANK_REASON This is a research paper detailing a new benchmark and evaluation of AI models. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- digital single-lens reflex camera
- EfficientNet
- Hugging Face
- Long-Range Video Presentation Attack Detection: A Three-Class Benchmark Across 1-100 m and Two Sensor Types
- MobileNet
- MobileNetV3
- ResNet-50
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →