3D-SPFES: A Monocular Depth-Aware Geometric Graph Neural Network for Sheep Facial Pain Assessment
A recent study presents the 3D Sheep Pain Facial Expression System (3D-SPFES), a monocular depth-aware geometric graph neural network aimed at evaluating pain in sheep through three-dimensional facial expression analysis. This innovative system incorporates landmarks from the established Sheep Pain Facial Expression Scale (SPFES)—including the ears, eyes, and nose—into a 3D Euclidean framework derived from a single RGB camera using VideoDepthAnything, thus negating the need for specialized depth equipment. Each landmark is represented by a feature vector that includes its 3D coordinates, estimated surface normal, and facial attribute class embedding. The connections between nodes are weighted using a metric that combines Euclidean distance and surface curvature. The paper, available on arXiv (reference 2608.11050), points out the shortcomings of conventional 2D deep learning systems, which oversimplify facial structures and overlook essential 3D anatomical relationships, ultimately enhancing the precision of pain evaluations in sheep and promoting better animal welfare in agricultural practices.
Key facts
- The paper introduces 3D-SPFES, a novel monocular depth-aware geometric graph neural network for sheep facial pain assessment.
- 3D-SPFES integrates SPFES facial landmarks into 3D Euclidean space using a single RGB camera and VideoDepthAnything.
- The system avoids the need for specialized depth hardware.
- Each landmark node includes 3D spatial coordinates, estimated surface normal, and facial attribute class embedding.
- Edges are weighted based on Euclidean distance and surface curvature.
- The paper criticizes traditional 2D deep learning systems for losing 3D anatomy and cross-landmark relationships.
- The research is published on arXiv with reference 2608.11050.
- The system is based on the clinically proven Sheep Pain Facial Expression Scale (SPFES).
Entities
Institutions
- arXiv