Papers › Shape Preserving Facial Landmarks with Graph Attention Networks

Shape Preserving Facial Landmarks with Graph Attention Networks

13 Oct 2022arXiv:2210.07233archive 2025-07-28

Andrés Prados-Torreblanca, José M. Buenaposada, Luis Baumela

Top-performing landmark estimation algorithms are based on exploiting the excellent ability of large convolutional neural networks (CNNs) to represent local appearance. However, it is well known that they can only learn weak spatial relationships. To address this problem, we propose a model based on the combination of a CNN with a cascade of Graph Attention Network regressors. To this end, we introduce an encoding that jointly represents the appearance and location of facial landmarks and an attention mechanism to weigh the information according to its reliability. This is combined with a multi-task approach to initialize the location of graph nodes and a coarse-to-fine landmark description scheme. Our experiments confirm that the proposed model learns a global representation of the structure of the face, achieving top performance in popular benchmarks on head pose and landmark estimation. The improvement provided by our model is most significant in situations involving large changes in the local appearance of landmarks.

PaperPDFCode

In Syntology View this paper on Syntology: its repositories, every harvested function with whether it ran, its licence and the call to fetch it.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

andresprados/spiga officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

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

Tasks

Face AlignmentFacial Landmark DetectionGraph AttentionHead Pose EstimationPose Estimation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Face Alignment 300W SPIGA NME_inter-ocular (%, Challenge) 4.66 #10 of 48 Archive leaderboard report
Face Alignment 300W SPIGA NME_inter-ocular (%, Common) 2.59 #10 of 48 Archive leaderboard report
Face Alignment 300W SPIGA NME_inter-ocular (%, Full) 2.99 #10 of 48 Archive leaderboard report
Face Alignment 300W SPIGA NME_inter-pupil (%, Challenge) 6.73 #10 of 48 Archive leaderboard report
Face Alignment 300W SPIGA NME_inter-pupil (%, Common) 3.59 #10 of 48 Archive leaderboard report
Face Alignment 300W SPIGA NME_inter-pupil (%, Full) 4.20 #10 of 48 Archive leaderboard report
Face Alignment 300W (Common) SPIGA NME 2.59 #1 of 2 Archive leaderboard report
Face Alignment 300W Split 2 SPIGA AUC@7 (box) 71.0 #1 of 7 Archive leaderboard report
Face Alignment 300W Split 2 SPIGA AUC@8 (inter-ocular) 57.27 #1 of 7 Archive leaderboard report
Face Alignment 300W Split 2 SPIGA FR@8 (inter-ocular) 0.67 #1 of 7 Archive leaderboard report
Face Alignment 300W Split 2 SPIGA NME (box) 2.03 #1 of 7 Archive leaderboard report
Face Alignment 300W Split 2 SPIGA NME (inter-ocular) 3.43 #1 of 7 Archive leaderboard report
Face Alignment COFW-68 SPIGA AUC@7 (box) 64.1 #1 of 7 Archive leaderboard report
Face Alignment COFW-68 SPIGA NME (box) 2.52 #1 of 7 Archive leaderboard report
Face Alignment COFW-68 SPIGA NME (inter-ocular) 3.93 #1 of 7 Archive leaderboard report
Face Alignment MERL-RAV SPIGA AUC@7 (box) 78.47 #1 of 2 Archive leaderboard report
Face Alignment MERL-RAV SPIGA NME (box) 1.51 #1 of 2 Archive leaderboard report
Face Alignment WFLW SPIGA AUC@10 (inter-ocular) 60.56 #6 of 36 Archive leaderboard report
Face Alignment WFLW SPIGA FR@10 (inter-ocular) 2.08 #6 of 36 Archive leaderboard report
Face Alignment WFLW SPIGA NME (inter-ocular) 4.06 #6 of 36 Archive leaderboard report
Face Alignment WFW (Extra Data) SPIGA AUC@10 (inter-ocular) 60.56 #3 of 11 Archive leaderboard report
Face Alignment WFW (Extra Data) SPIGA FR@10 (inter-ocular) 2.08 #3 of 11 Archive leaderboard report
Face Alignment WFW (Extra Data) SPIGA NME (inter-ocular) 4.06 #3 of 11 Archive leaderboard report
Facial Landmark Detection 300W SPIGA (Inter-ocular Norm) NME 2.99 #3 of 15 Archive leaderboard report
Head Pose Estimation WFLW SPIGA MAE mean (º) 1.52 #1 of 2 Archive leaderboard report
Head Pose Estimation WFLW SPIGA MAE pitch (º) 1.86 #1 of 2 Archive leaderboard report
Head Pose Estimation WFLW SPIGA MAE roll (º) 0.93 #1 of 2 Archive leaderboard report
Head Pose Estimation WFLW SPIGA MAE yaw (º) 1.78 #1 of 2 Archive leaderboard report
Pose Estimation 300W (Full) SPIGA MAE mean (º) 1.29 #1 of 3 Archive leaderboard report
Pose Estimation 300W (Full) SPIGA MAE pitch (º) 1.70 #1 of 3 Archive leaderboard report
Pose Estimation 300W (Full) SPIGA MAE roll (º) 0.77 #1 of 3 Archive leaderboard report
Pose Estimation 300W (Full) SPIGA MAE yaw (º) 1.41 #1 of 3 Archive leaderboard report
Pose Estimation MERL-RAV SPIGA MAE mean (º) 2.39 #1 of 1 Archive leaderboard report
Pose Estimation MERL-RAV SPIGA MAE pitch (º) 2.24 #1 of 1 Archive leaderboard report
Pose Estimation MERL-RAV SPIGA MAE roll (º) 1.71 #1 of 1 Archive leaderboard report
Pose Estimation MERL-RAV SPIGA MAE yaw (º) 3.23 #1 of 1 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.

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