Reproduce “ChangeT$^3$: Evaluating Vision Test-Time Training for Remote Sensing Change Detection” 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
Sharukshan Niranjan, Oshadha Samarakoon, Nichula Wasalathilaka, Roshan Godaliyadda, Parakrama Ekanayake, Vijitha Herath
ChangeT³ adapts hierarchical visual test-time training to paired satellite images for building change detection. It processes each date independently with temporary image-conditioned weights, then uses a multiscale decoder to compare features. On LEVIR-CD+, its change-class F1 averaged 80.61 ± 0.90%, versus 78.84 ± 0.21% for matched ChangeFormer. However, identical validation and test files prevent an independent test estimate.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.