VIBEFACE - Video and Image Biometric Dataset for Evaluation of Faces

ICLR 2026 Conference Submission18754 Authors

Keywords: Face, eKYC, Demografic balance, Face verification, Face detection

TL;DR: The paper introduces a balanced biometric dataset for face verification, comprising standardized and selfie photos, as well as selfie and eKYC videos.

Abstract: Face verification has become a critical component of modern security systems, with facial biometrics widely adopted in sectors such as banking, mobile device authentication, and secure access to applications. To develop and evaluate robust biometric systems, a comprehensive dataset is required. While existing facial biometric datasets often address variations in pose, lighting, and demographics, they rarely capture realistic operational settings such as electronic Know Your Client (eKYC) procedures — a critical requirement for financial and regulatory compliance. To address this gap, we present VIBEFACE, a novel dataset specifically designed to support face verification in eKYC and related scenarios. VIBEFACE comprises 2,250 high-quality facial images (1,250 standardized and 1,000 selfie photographs) and 1,550 short videos, including both selfie recordings and sequences that explicitly mimic eKYC workflows. Data were collected using mobile device cameras from 50 diverse subjects, ensuring coverage and balance of demographic attributes, including gender, race, and age. By integrating realistic eKYC sequences with diverse visual conditions and strict adherence to ethical and legal standards, VIBEFACE establishes a new benchmark for evaluating the robustness and fairness of biometric verification systems in practical, compliance-driven environments.

Primary Area: datasets and benchmarks

Submission Number: 18754