Phone and Webcam Video Dataset - 30,952 files for Anti-spoofing and Facial Recognition

Overview
The Phone and Webcam Video Dataset is a large-scale multimodal biometric collection produced by Unidata, designed to support research in face recognition, liveness detection, anti-spoofing, and biometric verification. The dataset comprises 30,952 files from 3,869 individuals across 16 countries, recorded using both mobile phones and webcams under real-world conditions.
Each participant completed a standardized recording protocol — pronouncing a sequence of numbers — while contributing 8 files per person: 4 photos and 4 videos captured across two device types. This structured, dual-device approach makes the dataset especially well-suited for cross-device generalization studies and training robust detection algorithms.

Dataset Structure
Each participant provides a full set of 8 files:
1. Photos (4 total):
- 2 photos captured on a mobile phone
- 2 photos captured on a webcam
2. Videos (4 total):
- 2 videos captured on a mobile phone (one ~30 seconds, one ~8 seconds)
- 2 videos captured on a webcam (same durations)
All video files are provided in MP4 format. The recording task — pronouncing a set of numbers — introduces a liveness signal directly into the capture protocol, supporting challenge-response anti-spoofing pipelines out of the box.

Subject Demographics
Total individuals: 3,869 across 17 countries
Gender: male and female participants
Ethnicity: African and Indonesian
Age distribution:
- Under 18: 98 people
- 19–25: 1,527 people
- 26–32: 983 people
- 33–39: 723 people
- 40–46: 278 people
- 47–53: 159 people
- 54–59: 50 people
- 60–66: 32 people
- 67+: 19 people
Data was collected via a crowdsourcing platform by the Unidata team.

Technical Specifications
- Video format: MP4
- Video durations: ~30 seconds (long take), ~8 seconds (short take)
- Phone models: Vivo, Realme, Oppo, Tecno, Xiaomi, Infinix, and others
- Webcam models: Logitech, Havit, Xiaomi, VerkTop, Aukey, and others
- Labeling / metadata per subject: ID, gender, ethnicity, phone type, webcam type
The dataset includes diverse hardware across both smartphone cameras and webcam streams, reflecting real deployment conditions for mobile and desktop biometric systems.

Use Cases
- Face recognition & re-identification — train and evaluate models that identify individuals across different devices, sessions, and recording conditions
- Anti-spoofing & liveness detection — the verbal prompt protocol and dual-device capture create natural variation useful for detecting presentation attacks in video streams
- Biometric verification — paired phone and webcam recordings from the same subjects support development and testing of authentication systems
Object detection & face detection — high-quality videos with consistent framing and known subject metadata enable benchmark evaluation of detection pipelines
- Cross-device generalization — recordings on both Android phones and webcams from the same individuals allow direct study of domain shift between device types
- Training data for learning models — scale, device diversity, and structured annotations make the dataset a reliable foundation for computer vision and machine learning workflows

Compliance & Access
All data was collected under informed consent and complies with GDPR and applicable data protection regulations. Storage is hosted on AWS, certified to ISO 27001 and ISO 27701 standards.

Summary
The Phone and Webcam Video Dataset is a large-scale, dual-device biometric collection purpose-built for training and evaluating face recognition, liveness detection, and anti-spoofing systems. Its core strength lies in the structured 8-file recording protocol — 4 photos and 4 videos per subject, captured on both mobile phones and webcams — which directly mirrors the device diversity found in real-world biometric deployments.
With 30,952 files from 3,869 individuals spanning 17 countries, two ethnicities (African and Indonesian), and a full adult age range from under 18 to 67+, the dataset provides the demographic breadth needed to build models that generalize beyond controlled lab conditions. Support for a wide range of consumer devices — including Vivo, Xiaomi, Oppo, Realme smartphones and Logitech, Aukey, Havit webcams — further ensures that trained models are robust to sensor variation across Android phones and webcam streams.