VIBEFACE - Video and Image Biometric Dataset

Dataset Overview

A comprehensive facial biometric dataset comprising 2,250 high-quality images and 1,550 videos from 50 demographically balanced subjects, designed for fair and robust evaluation of face recognition systems under realistic conditions.

120

Subjects

5,400

Images

3,720

Videos

Scenarios

Sample Faces

Demographics & Balance

Equal representation across gender, race, and age groups ensures fair evaluation of biometric systems.

Dataset	IDs	Photos	Videos	eKYC	Glasses	DD
MOBIO	150		✔️
Replay-Mobile	40	✔️	✔️
OULU-NPU	55		✔️
MobiBits	53	✔️			✔️	✔️
WMCA	72		✔️		✔️
HQ-WMCA	51		✔️		✔️
Soteria	70	✔️	✔️		✔️	✔️
eKYC-DF	100		✔️	✔️		✔️
VIBEFACE (Ours)	120	✔️	✔️	✔️	✔️	✔️

Session Capture Conditions

Session	Light conditions	Glasses	Scenarios
			Standardized photos (Scenarios 1–5) Back camera	Selfie photos (Scenarios 6–10) Front camera	Selfie video (Scenario 11) Front camera	Verification videos (Scenarios 12–18) Front camera
A	Artificial light	No glasses	+	+	+	+
B	Flash	No glasses	+	-	–	–
C	Artificial light	Glasses	+	+	-	+
D	Natural light	No glasses	+	+	+	+
E	Weak natural light	No glasses	+	+	+	+

Interactive Biometric Analysis

Similarity Timeline

Frame: 0 Time: 0.00s Total: --

ArcFace

0.000

MagFace

0.000

Head Pose

Pitch

0°

Yaw

0°

Roll

0°

Benchmark Results

Comprehensive evaluation across face detection, verification, age estimation, and gender classification tasks.

IDENTIFICATION BENCHMARKS (click to expand)

AGE/GENDER ESTIMATION BENCHMARKS (click to expand)

We are using the middle value from the FairFace model for age estimation

FACE VERIFICATION BENCHMARKS (click to expand)

Face verification performance broken down by scenario (Scn), gender, age group, and racial category.
Abbreviations: Afr. – African, Cauc. – Caucasian, EA – East Asian, SA – South Asian.

Scn.	Session				Gender		Age group			Ethnic group				All
Scn.	A	C	D	E	Female	Male	18-30	31-50	51-70	Afr.	Cauc.	EA	SA	All
ArcFace
OAV	0.473	0.390	0.467	0.455	0.424	0.439	0.419	0.438	0.440	0.443	0.412	0.411	0.477	0.431
FV	0.854	0.735	0.843	0.838	0.830	0.833	0.831	0.836	0.826	0.827	0.826	0.838	0.841	0.832
12	0.810	0.688	0.801	0.791	0.778	0.767	0.775	0.781	0.757	0.776	0.773	0.763	0.780	0.772
13	0.792	0.676	0.805	0.778	0.763	0.763	0.763	0.769	0.755	0.766	0.750	0.765	0.777	0.763
14	0.836	0.707	0.849	0.838	0.805	0.809	0.815	0.816	0.785	0.808	0.794	0.818	0.816	0.807
15	0.831	0.712	0.848	0.843	0.806	0.811	0.812	0.818	0.790	0.805	0.800	0.816	0.819	0.809
16	0.823	0.699	0.837	0.833	0.801	0.795	0.803	0.810	0.774	0.792	0.784	0.814	0.810	0.798
17	0.835	0.726	0.843	0.847	0.813	0.813	0.818	0.819	0.797	0.810	0.801	0.819	0.830	0.813
18	0.833	0.726	0.842	0.844	0.810	0.812	0.816	0.819	0.794	0.810	0.799	0.816	0.830	0.811
MagFace
OAV	0.326	0.290	0.328	0.325	0.296	0.300	0.281	0.306	0.312	0.291	0.290	0.290	0.334	0.298
FV	0.817	0.723	0.796	0.800	0.805	0.796	0.803	0.806	0.791	0.795	0.784	0.818	0.815	0.801
12	0.763	0.666	0.761	0.756	0.746	0.727	0.735	0.747	0.724	0.732	0.735	0.739	0.742	0.736
13	0.750	0.650	0.770	0.748	0.735	0.724	0.731	0.733	0.722	0.720	0.719	0.741	0.745	0.729
14	0.800	0.705	0.821	0.817	0.790	0.781	0.796	0.790	0.765	0.779	0.765	0.810	0.797	0.786
15	0.792	0.700	0.815	0.812	0.785	0.774	0.784	0.788	0.761	0.764	0.766	0.803	0.795	0.780
16	0.780	0.684	0.803	0.801	0.774	0.760	0.774	0.773	0.748	0.754	0.753	0.790	0.779	0.767
17	0.798	0.722	0.813	0.825	0.795	0.784	0.798	0.789	0.779	0.776	0.773	0.813	0.809	0.790
18	0.794	0.717	0.809	0.821	0.792	0.779	0.793	0.786	0.774	0.779	0.767	0.803	0.803	0.785

False Rejection Rate comparison across different racial groups and testing conditions using ArcFace and MagFace models.

Half-profile and frontal images were used.
Click on legend to disable color.

ArcFace FRR by Race

MagFace FRR by Race

Overall FRR Comparison

Access Dataset

Available for research purposes under controlled-access agreement with full GDPR compliance.

Download dataset