Using ViT-H and ArcFace to Train a Head Model for Universal Image Embedding - 42Papers