FGTBT: Frequency-guided task-balancing transformer for unified facial landmark detection

2026-01-01

Information Sciences

AI Summary

Existing facial landmark detection methods struggle with modeling geometric structures and lack robustness under challenging conditions such as large head poses, extreme illumination variations, and diverse facial expressions. To address these limitations, this work proposes a frequency-guided task-balanced Transformer framework. The approach introduces a Frequency-Guided Structure-Aware (FGSA) module to incorporate structural priors from the frequency domain and designs a Fine-grained Multi-task Balanced loss (FMB-loss) that enables landmark-level adaptive weighting to mitigate gradient conflicts among multiple tasks. Combined with a unified cross-dataset training strategy, the proposed method achieves state-of-the-art performance on several mainstream benchmarks, significantly improving both accuracy and robustness in complex real-world scenarios.

Technology Category

Computer Vision: Biometrics, Face, Gesture & PoseMachine Learning: Transfer, Domain Adaptation, Multi-Task LearningNatural Language Processing: Language Grounding & Multi-modal NLP

Application Category

Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphsSearch and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAGWeb Mining and Content Analysis: Large pretrained models with web data

Problem

• facial landmark detection
• pose variations
• illumination changes
• expression variations
• dataset diversity

Innovation

• Frequency-Guided
• Task-Balancing
• Facial Landmark Detection
• Multi-Task Learning
• Transformer