PaCo-FR: Patch-Pixel Aligned End-to-End Codebook Learning for Facial Representation Pre-training

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xie, Yin, Chen, Zhichao, Xiao, Zeyu, Zhao, Yongle, An, Xiang, Yang, Kaicheng, Ran, Zimin, Guo, Jia, Feng, Ziyong, Deng, Jiankang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908940334268416
author Xie, Yin
Chen, Zhichao
Xiao, Zeyu
Zhao, Yongle
An, Xiang
Yang, Kaicheng
Ran, Zimin
Guo, Jia
Feng, Ziyong
Deng, Jiankang
author_facet Xie, Yin
Chen, Zhichao
Xiao, Zeyu
Zhao, Yongle
An, Xiang
Yang, Kaicheng
Ran, Zimin
Guo, Jia
Feng, Ziyong
Deng, Jiankang
contents Facial representation pre-training is crucial for tasks like facial recognition, expression analysis, and virtual reality. However, existing methods face three key challenges: (1) failing to capture distinct facial features and fine-grained semantics, (2) ignoring the spatial structure inherent to facial anatomy, and (3) inefficiently utilizing limited labeled data. To overcome these, we introduce PaCo-FR, an unsupervised framework that combines masked image modeling with patch-pixel alignment. Our approach integrates three innovative components: (1) a structured masking strategy that preserves spatial coherence by aligning with semantically meaningful facial regions, (2) a novel patch-based codebook that enhances feature discrimination with multiple candidate tokens, and (3) spatial consistency constraints that preserve geometric relationships between facial components. PaCo-FR achieves state-of-the-art performance across several facial analysis tasks with just 2 million unlabeled images for pre-training. Our method demonstrates significant improvements, particularly in scenarios with varying poses, occlusions, and lighting conditions. We believe this work advances facial representation learning and offers a scalable, efficient solution that reduces reliance on expensive annotated datasets, driving more effective facial analysis systems.
format Preprint
id arxiv_https___arxiv_org_abs_2508_09691
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PaCo-FR: Patch-Pixel Aligned End-to-End Codebook Learning for Facial Representation Pre-training
Xie, Yin
Chen, Zhichao
Xiao, Zeyu
Zhao, Yongle
An, Xiang
Yang, Kaicheng
Ran, Zimin
Guo, Jia
Feng, Ziyong
Deng, Jiankang
Computer Vision and Pattern Recognition
Facial representation pre-training is crucial for tasks like facial recognition, expression analysis, and virtual reality. However, existing methods face three key challenges: (1) failing to capture distinct facial features and fine-grained semantics, (2) ignoring the spatial structure inherent to facial anatomy, and (3) inefficiently utilizing limited labeled data. To overcome these, we introduce PaCo-FR, an unsupervised framework that combines masked image modeling with patch-pixel alignment. Our approach integrates three innovative components: (1) a structured masking strategy that preserves spatial coherence by aligning with semantically meaningful facial regions, (2) a novel patch-based codebook that enhances feature discrimination with multiple candidate tokens, and (3) spatial consistency constraints that preserve geometric relationships between facial components. PaCo-FR achieves state-of-the-art performance across several facial analysis tasks with just 2 million unlabeled images for pre-training. Our method demonstrates significant improvements, particularly in scenarios with varying poses, occlusions, and lighting conditions. We believe this work advances facial representation learning and offers a scalable, efficient solution that reduces reliance on expensive annotated datasets, driving more effective facial analysis systems.
title PaCo-FR: Patch-Pixel Aligned End-to-End Codebook Learning for Facial Representation Pre-training
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.09691