Labeled Faces in the Wild Dataset Used in “Robust Tensor-on-Tensor Regression” (Hirari et al., 2026)

The Labeled Faces in the Wild (LFW) dataset is a publicly available collection of face images gathered from the web. It contains more than 13,000 images of individuals captured under unconstrained conditions, exhibiting substantial variability in lighting, pose, image quality, and background. Each image is associated with a set of 72 continuous attributes describing facial characteristics such as gender, age, expression, and accessories. In our analysis, we consider a subset of 400 images corresponding to distinct individuals. Each image is converted to grayscale and downsampled to a resolution of 30x30 pixels, resulting in tensor-valued predictors of dimension 30x30. The associated attribute vectors serve as multivariate responses. This preprocessing yields a dataset suitable for tensor-on-tensor regression, where the goal is to predict facial attributes from image data.This dataset is used in the empirical study of Hirari, M., Centofanti, F., Hubert, M., and Van Aelst, S. (2026), Robust Tensor-on-Tensor Regression.