Real-world attack datasets

Custom data collection for biometric systems, focused on real attack scenarios and production-level conditions

Discuss your case

built around your system — not generic data

We focus on collecting data for specific failure scenarios and real-world conditions relevant to your system

Focus on attack conditions and edge cases

Designed for models improvement

Adapted to real system behavior

Scenario-driven data collection

[00%]

Problems & challenges

Typical challenges we help solve

  • Weakness on edge cases

    Model performance drops on rare, complex, or atypical real-world scenarios

  • Limited attack coverage

    The dataset includes insufficient attack scenarios, reducing robustness against adversarial or harmful inputs

  • Misaligned training data

    Training data does not accurately reflect real production behavior or user interactions

  • Production-aligned data

    Data collection is designed around your production pipeline and mirrors real user behavior

  • Environment & device specific

    Data is collected across relevant environments and devices to match real usage conditions

Process

How we work

  • Define target scenarios

    We align on specific cases and conditions relevant to your system

  • Design data collection approach

    We structure the data collection around real-world usage and attack scenarios

  • Execute data collection

    Data is collected across controlled and variable conditions

  • Deliver structured dataset

    Data is prepared for direct use in ML pipelines

Main directions

What we work with

  • Face anti-spoofing

    Data collection for presentation attack scenarios under diverse real-world conditions

    • Print attack
    • Replay attack
    • Cutout attack
    • Environment variability
    • Device variability
  • Synthetic scenarios & injection

    Simulation of digital attack vectors affecting system inputs

    • injection attack
    • Virtual camera Variability
    • OS Variability
    • Browser Variability
  • Templates & fraud data

    Data supporting document-based systems and fraud scenarios

    • Print attack
    • Replay attack
    • Templates Variability
    • Cutout attack
    • Environment variability
    • Device variability
  • Voice

    Datasets for synthetic speech conditions

    • Replay attack
    • Synthetic replay
    • Device variability
    • Distance Variability