Reproduce “A Unified Objective for Novel Class Discovery” 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
Enrico Fini, Enver Sangineto, Stéphane Lathuilière, Zhun Zhong, Moin Nabi, Elisa Ricci
UNO trains known and novel image categories within one prediction problem, combining ground-truth labels for known classes with pseudo-labels for unlabeled images in a shared classification objective. This makes supervised recognition and novel-class discovery cooperate. On ImageNet, its best head reaches 90.6% average clustering accuracy on the unlabeled training split, versus 82.5% for RS+, under task-aware evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.