Papers › Fiducial Focus Augmentation for Facial Landmark Detection

Fiducial Focus Augmentation for Facial Landmark Detection

23 Feb 2024arXiv:2402.15044archive 2025-07-28

Purbayan Kar, Vishal Chudasama, Naoyuki Onoe, Pankaj Wasnik, Vineeth Balasubramanian

Deep learning methods have led to significant improvements in the performance on the facial landmark detection (FLD) task. However, detecting landmarks in challenging settings, such as head pose changes, exaggerated expressions, or uneven illumination, continue to remain a challenge due to high variability and insufficient samples. This inadequacy can be attributed to the model's inability to effectively acquire appropriate facial structure information from the input images. To address this, we propose a novel image augmentation technique specifically designed for the FLD task to enhance the model's understanding of facial structures. To effectively utilize the newly proposed augmentation technique, we employ a Siamese architecture-based training mechanism with a Deep Canonical Correlation Analysis (DCCA)-based loss to achieve collective learning of high-level feature representations from two different views of the input images. Furthermore, we employ a Transformer + CNN-based network with a custom hourglass module as the robust backbone for the Siamese framework. Extensive experiments show that our approach outperforms multiple state-of-the-art approaches across various benchmark datasets.

PaperPDF

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Face AlignmentFacial Landmark DetectionImage Augmentation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Face Alignment 300W FiFA NME_inter-ocular (%, Challenge) 4.47 #4 of 48 Archive leaderboard report
Face Alignment 300W FiFA NME_inter-ocular (%, Common) 2.51 #4 of 48 Archive leaderboard report
Face Alignment 300W FiFA NME_inter-ocular (%, Full) 2.89 #4 of 48 Archive leaderboard report
Face Alignment AFLW-19 FiFA AUC_box@0.07 (%, Full) 81.8 #1 of 23 Archive leaderboard report
Face Alignment AFLW-19 FiFA NME_box (%, Full) 1.31 #1 of 23 Archive leaderboard report
Face Alignment AFLW-19 FiFA NME_diag (%, Frontal) 0.80 #1 of 23 Archive leaderboard report
Face Alignment AFLW-19 FiFA NME_diag (%, Full) 0.92 #1 of 23 Archive leaderboard report
Face Alignment COFW FiFA NME (inter-ocular) 2.96 #1 of 28 Archive leaderboard report
Facial Landmark Detection 300W FiFA NME 2.89 #2 of 15 Archive leaderboard report
Facial Landmark Detection AFLW-Front FiFA Mean NME 0.80 #1 of 3 Archive leaderboard report
Facial Landmark Detection AFLW-Front FiFA Mean NME 0.80 #1 of 3 Archive leaderboard report
Facial Landmark Detection AFLW-Front FiFA NME 0.80 #1 of 3 Archive leaderboard report
Facial Landmark Detection AFLW-Full FiFA Mean NME 0.92 #1 of 5 Archive leaderboard report
Facial Landmark Detection AFLW-Full FiFA Mean NME 0.92 #1 of 5 Archive leaderboard report
Facial Landmark Detection AFLW-Full FiFA NME 0.92 #1 of 5 Archive leaderboard report
Facial Landmark Detection COFW FiFA NME 2.96 #2 of 2 Archive leaderboard report
Facial Landmark Detection COFW FiFA NME (inter-ocular) 2.96 #2 of 2 Archive leaderboard report
Facial Landmark Detection WFLW FiFA AUC@10 (inter-ocular) 61.78 #3 of 3 Archive leaderboard report
Facial Landmark Detection WFLW FiFA FR@10 (inter-ocular) 1.60 #3 of 3 Archive leaderboard report
Facial Landmark Detection WFLW FiFA NME (inter-ocular) 3.89 #3 of 3 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

1x1 ConvolutionAbsolute Position EncodingsAdamAttentionBPEConvolutionDense ConnectionsDropoutHourglass ModuleLabel SmoothingLayer NormalizationLinear LayerMax PoolingMulti-Head AttentionPosition-Wise Feed-Forward LayerReLUResidual ConnectionSoftmaxTransformer

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections