SlackedFace: Learning a Slacked Margin for Low-Resolution Face Recognition

Cheng Yaw Low (Institute for Basic Science),* Jacky Chen Long Chai (Yonsei University), Jaewoo Park (Yonsei University), KYEONGJIN ANN (KAIST), Meeyoung Cha (KAIST & IBS)
The 34th British Machine Vision Conference


Low-resolution (LR) face recognition poses a significant challenge in embedding learning due to the severe loss of identity information. Recent softmax losses introduce a non-static margin that gives greater importance or a larger margin to recognizable examples based on the embedding norm as a measure of face recognizability. In this paper, we argue that face recognizability is more than just the embedding norm, as it does not capture the fine-grained details of face images that are critical to embedding learning. We propose SlackedFace to induce a relaxed margin aligned with face recognizability and the model's confidence based on both embedding norm and embedding proximity for empowered embedding learning. We also introduce fast-hill climbing as an early calibration stage between pre-trained and randomly initialized modules. We show that SlackedFace outperforms the current best models on realistic LR face datasets when tested in practical open-set evaluation scenarios.



author    = {Cheng Yaw Low and Jacky Chen Long Chai  and Jaewoo Park and KYEONGJIN ANN and Meeyoung Cha},
title     = {SlackedFace: Learning a Slacked Margin for Low-Resolution Face Recognition},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year      = {2023},
url       = {}

Copyright © 2023 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection