FACES : facial analysis with compressed efficient systems

Lajić, Romanela (avtor), Peer, Peter (avtor), Štruc, Vitomir (avtor), Han, Dong Seog (avtor), Meden, Blaž (avtor), Emeršič, Žiga (avtor)

Due to their promising performance vision transformers are increasingly being incorporated into various biometric solutions, mainly in the domain of face analysis. However, their size and computational expense remain the biggest challenge when it comes to their full utilization and there is a high demand for optimization of these models. In this paper we propose a novel pruning technique for face analysis vision transformers aimed at reducing their memory and computational cost. The method uses existing transformer parameters as importance scores, which allows for a simple one-shot pruning and retraining approach. By testing the method on the SWINFace transformer for both verification and attribute recognition tasks, we show that the models compressed up to 50% sparsity level maintain the performance or even outperform the original model, while also outperforming state-of-the-art vision transformer pruning methods and showing versatility for different face analysis tasks.

• Jezik: Angleški jezik
• Ključne besede: face analysis, neural network compression, pruning, vision transformers
• Vrsta gradiva: Članek v reviji
• Tipologija: 1.01 - Izvirni znanstveni članek
• Organizacija: FRI - Fakulteta za računalništvo in informatiko
• Status publikacije: Objavljeno
• Različica publikacije: Objavljena publikacija
• Leto izida: 2026
• Št. strani: 5 str.
• Številčenje: Vol. , no.
• PID: 20.500.12556/RUL-181219
• UDK: 004.93:57.087.1
• ISSN pri članku: 2405-9595
• DOI: 10.1016/j.icte.2026.02.008
• COBISS.SI-ID: 273178115
• Datum objave v RUL: 27.03.2026

Gradivo je del revije

• Naslov: ICT express
• Založnik: Elsevier
• ISSN: 2405-9595
• COBISS.SI-ID: 526132505

Licence

• Licenca: CC BY-NC-ND 4.0, Creative Commons Priznanje avtorstva-Nekomercialno-Brez predelav 4.0 Mednarodna
• Povezava: http://creativecommons.org/licenses/by-nc-nd/4.0/deed.sl
• Opis: Najbolj omejujoča licenca Creative Commons. Uporabniki lahko prenesejo in delijo delo v nekomercialne namene in ga ne smejo uporabiti za nobene druge namene.

Sekundarni jezik

• Jezik: Slovenski jezik
• Ključne besede: analiza obraza, kompresija nevronskih mrež, obrezovanje, vidni transformatorji, ViT

Projekti

• Financer: ARIS - Javna agencija za znanstvenoraziskovalno in inovacijsko dejavnost Republike Slovenije
• Številka projekta: P2-0250-2018
• Naslov: Metrologija in biometrični sistemi

• Financer: EC - European Commission
• Številka projekta: 101225635
• Naslov: On-the-move Schengen border control using extended EUDI wallet, smartphone and external sensor technologies
• Akronim: OnMoveID