Reproduce “Equivariant Similarity for Vision-Language Foundation Models” 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
Tan Wang, Kevin Lin, Linjie Li, Chung-Ching Lin, Zhengyuan Yang, Hanwang Zhang, Zicheng Liu, Lijuan Wang
EqSim links similarity changes across matched image-text pairs through regularization, using stronger constraints for model-scored close examples and simpler constraints for distant ones. It also introduces EqBen, a benchmark of subtle visual differences. On EqBen, FIBER’s average rose from 43.40 to 45.32 after Flickr30K fine-tuning with EqSim, averaging 15 scores; its EQ-SD group score fell from 81.29 to 80.70.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.