Reproduce “Behavioral Affinity Transduction for Zero-Shot Vision-Language Classification” 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
Junxi Huang, Wenshuo Chen, Xiucheng Wang, Meitong Shen, Shaofeng Liang, Yuehan Kou, Yutao Yue, Nan Cheng
The authors introduce a plug-in for frozen vision-language models that uses class-score profiles to connect images in an unlabeled batch, then incorporates neighbors’ class assignments into prediction. On a ten-dataset average, soft and hard EM-Dirichlet accuracy rise from 68.99% to 73.21% and from 70.69% to 73.41%, respectively. This is transductive inference on a single batch, not independent single-image prediction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.