Preprocessing-impact study for face-shape classification: 56 training runs across 4 CNN/ViT models x 4 data-preparation pipelines, with seed replications, an alignment ablation, and near-duplicate/identity leakage audits. Data-availability artifact for the paper.
computer-vision reproducible-research pytorch transfer-learning data-preparation image-preprocessing vision-transformer face-shape-classification
-
Updated
Oct 31, 2026 - Python