Reproduce “SARATR-X-v2: Scale-Aware Structural Pre-Training for SAR 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
Weijie Li, Yafei Song, Yongxiang Liu, Bowen Peng, Jie Zhou, Jingyuan Xia, Wei Yang, Tianpeng Liu and 2 more
SARATR-X-v2 pre-trains SAR image representations by reconstructing a learned mixture of fixed structural measurements at six scales, rather than raw pixels or separately supervised scales. Its target is designed to be more stable under speckle. Across 12 transfer benchmarks, it ranks first on 10 and second on two; synthetic-speckle testing shows lower target drift, not general robustness to real-world shifts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.