Reproduce “Contrastive Embedding for Generalized Zero-Shot Learning” 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
Zongyan Han, Zhenyong Fu, Shuo Chen, Jian Yang
CE-GZSL addresses generalized zero-shot recognition by generating visual features from class descriptions, then embedding generated and real seen-class features in a space trained with instance-level and class-level contrastive signals. A final classifier learns from both feature sets. On CUB, it reports unseen/seen accuracies of 63.9/66.8 and harmonic mean 65.3, under fixed ResNet-101 features and proposed splits.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.