Reproduce “Supervised Metric Learning to Rank for Retrieval via Contextual Similarity Optimization” 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
Christopher Liao, Theodoros Tsiligkaridis, Brian Kulis
The authors introduce contextual similarity for supervised image retrieval, training image embeddings using overlap between images’ nearest-neighbor sets alongside same-label targets. The loss adds group-level relationships to pairwise supervision, with a heuristic gradient for discrete neighbor membership. On CUB, a 512-dimensional ResNet-50 embedding reaches 69.8 ± 0.2 R@1, versus ROADMAP’s 68.7 ± 0.5, but robustness tests use contextual loss alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.