NuTonic/sat-image-boundingbox-sft

NU-TONIC raw SFT init数据集包含卫星图像和对应的土地覆盖标签，以图像-文本行的形式打包，用于微调LFM-VL模型。数据来源包括GeoGuessr风格的POIs、多光谱光学COGs（蓝/绿/红或视觉预览）、地球引擎的土地覆盖标签（重新投影到与RGB堆栈相同的10米网格，并通过最近邻下采样与掩码结合）。数据集结构包括RGB芯片、可选边界框可视化、每个POI的静态地图背景、训练/验证/测试分割以及元数据。数据集构建脚本和许可证信息也在README中提及。

NU-TONIC raw SFT init contains satellite imagery and aligned land-cover outputs packaged as image–text rows for fine-tuning LFM-VL (leap-finetune `vlm_sft` format). Data provenance includes GeoGuessr-style POIs, multispectral optical COGs (blue/green/red or visual preview), and land-cover rasters from Earth Engine (reprojected to the same 10 m grid as the RGB stack, then tiled and nearest-neighbor downsampled with the mask). Dataset layout includes RGB chips, optional bbox visualizations, overhead context stills per POI, train/validation/test splits, and metadata. Build scripts and licensing information are also provided in the README.