← back to paper
arxiv: 2607.27087 · 2 revisions
Step-Attention Refinement of DINOv3 Features for Efficient Anterior Eye Segmentation