Multimodal Implicit Neural Field and Spatio-Temporal Graph Attention for Personalized Smart Sportswear Comfort Prediction

Persistent challenges in smart sportswear—spatial scale mismatches across modalities, weak semantic alignment, limited pressure-distribution accuracy and poor generalization to individual body shapes—are addressed through a unified framework combining a multimodal neural field with a spatio-temporal graph attention network (ST-GAT). This method constructs a multimodal neural field to encode images, pressure maps, and posture as spatially continuous functions, achieving high-dimensional semantic alignment in a unified latent space. Furthermore, the ST-GAT captures the temporal evolution of pressure under the topological structure of the human body.