← back to paper
arxiv: 2607.07135 · 2 revisions
Sparse Attention for Dense Open-Vocabulary Prediction in CLIP