Reproduce “RS-OPSD: Reliable Privileged On-Policy-Self-Distillation for Ultra-High-Resolution Remote Sensing VQA” 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
Chengjie Jiang, Yunqi Zhou, Jiafeng Yan, Sihang Zhao, Chun Yuan, Jing Li
RS-OPSD trains a full-image model to answer fine-detail questions about huge aerial images by using targeted evidence and contextual zooms during training, then answering without visual search or tool calls. Ground-truth regions and answers support its reliability filtering. It scores 53.1, 61.5, and 33.3 on three benchmarks, averaging 49.3, four points above ZoomSearch in this comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.