Reproduce “Visual Recognition with Deep Nearest Centroids” 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
Wenguan Wang, Cheng Han, Tianfei Zhou, Dongfang Liu
DNC replaces a learned class-weight vector with a comparison between query features and multiple class sub-centroids. It clusters within-class features and assigns each query to its nearest representative. On ADE20K validation, DNC-FCN with ResNet101 scores 41.1% mIoU, compared with 39.9% for FCN using the same backbone. This supports the approach in this setup; segmentation still requires target-task training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.