IJCB-AFMFR 2026: Competition on Adapting Foundation Models for Face Recognition Using Synthetic Training Data

2026-07-27

AI Summary

This study systematically investigates the effectiveness of purely synthetic identity data for adapting foundation models to enhance face recognition performance under resource-constrained conditions. Building upon the CLIP ViT-L/14 architecture, the work establishes two evaluation tracks—full-data and limited-data—employing full fine-tuning with Sub-Center ArcFace and low-rank adaptation via LoRA, respectively, alongside synthetically generated identities produced by IDPERTURB. The research presents the first validation of synthetic-only data efficacy in face recognition, introduces a fairness-aware evaluation framework, and identifies optimal adaptation strategies across varying data scales. The adapted models significantly outperform the original CLIP, surpassing established baselines on benchmarks including LFW, IJB-B/C, and TinyFace, with DMSTI-Neurotechnology and Idiap-BSP achieving top rankings in their respective tracks.

Abstract

This paper presents a summary of the Competition on Adapting Foundation Models for Face Recognition Using Synthetic Training Data (AFMFR), held at the 2026 International Joint Conference on Biometrics (IJCB 2026). The competition received a total of eight valid submissions from four distinct teams across two complementary tracks: a Full Data Track, in which participants adapt the CLIP ViT-L/14 foundation model using large-scale synthetic identity data, and a Limited Data Track, designed to reflect more resource-constrained adaptation regimes. All training data was generated exclusively using IDPERTURB. Submitted solutions are ranked based on verification and identification performance across a diverse suite of benchmarks, including LFW, CFP-FP, AgeDB-30, CALFW, CPLFW, IJB-B, IJB-C, and TinyFace, using the Borda count method. Fairness evaluation is additionally conducted on the RFW dataset across four demographic groups. The results demonstrate that adaptation of the CLIP foundation model with synthetic training data substantially improves over the off-the-shelf model and, in several cases, surpasses the baseline. Notably, full fine-tuning with Sub-Center ArcFace (DMSTI-Neurotechnology) leads the Full Data Track, while rank-stabilized LoRA adaptation (Idiap-BSP) proves most effective under limited-data conditions.

Problem

• Face Recognition
• Foundation Models
• Synthetic Data
• Model Adaptation
• Biometrics

Innovation

• synthetic training data
• foundation model adaptation
• CLIP fine-tuning
• LoRA
• face recognition