Reproduce “The Earth in One Gaze: Training-Free Active Focus for UHR Remote Sensing Understanding” 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
Yao Zhang, Pengyu Dai, Wei Guo, Jian Liang, Jian Song, Yafei Ou, Hongruixuan Chen, Naoto Yokoya
GazeEarth helps a frozen multimodal model answer questions about tiny features in huge satellite images by requesting up to two grid cells, then deterministically warping their neighborhood into one continuous full-scene view. On three benchmarks with Qwen3-VL-8B, it reaches 55.65% mean accuracy, versus 51.38% for Overview and 47.31% direct answering. Gains over Overview are not significant on LRS-GRO.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.