Reproduce “Topology-Aware Representation Alignment for Semi-Supervised Vision-Language 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
Junwon You, Mihyun Jang, Sangwoo Mo, Jae-Hun Jung
ToMA adds a structural alignment objective to semi-supervised vision-language adaptation, aligning directions of selected image and text embedding relations built from minimum spanning trees and cycle-closing edges. On remote-sensing zero-shot classification, it reaches 83.1 ± 1.4% top-1 accuracy on RSICD-CLS, versus SemiCLIP’s 80.1 ± 1.4%, when labeled pairs and unlabeled images come from different distributions; this is task-specific evidence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.