Reproduce “Generalized Decoupled Learning for Enhancing Open-Vocabulary Dense Perception” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junjie Wang, Keyu Chen, Yulin Li, Bin Chen, Hengshuang Zhao, Xiaojuan Qi, Zhuotao Tian
DeCLIP fine-tunes CLIP representations for dense perception, labeling regions and pixels with text-described categories. It separates learning local visual meaning from spatial relationships, using distinct feature paths and no text queries or annotations for representation fine-tuning. In evaluations, decoupled learning raises COCO Thing region-classification mean accuracy from 69.5 baseline to 75.0 while improving segmentation; these results do not establish universal performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.