ChrisRPL/blackline-atlas-benchmark-seeds

Blackline Atlas民用设施中断基准种子是一个用于结构化民用设施中断分类的小型公共基准测试数据集。该数据集包含三个子集：Blackline内部公共种子（1行）、xBD公共种子（4行）和SpaceNet 8公共种子（4行）。数据集的主要目的是比较不同提示基线和小型视觉语言模型在相同可运行公共切片上的表现，并在Blackline适配前后回归检查结构化JSON行为。数据集布局包括可运行的基准行、使用的图像对和可用的公共来源文件。数据集适用于基准测试和演示材料、跨模型比较以及评估堆栈中狭窄部分的公共可重复性。Blackline Atlas Civilian Disruption Benchmark Seeds is a small public benchmark dataset for structured civilian disruption triage. The dataset includes three subsets: Blackline internal public seed (1 row), xBD public seed (4 rows), and SpaceNet 8 public seed (4 rows). The main purpose of the dataset is to compare prompt baselines and small VLMs on the same runnable public slices and regression-check structured JSON behavior before and after Blackline adaptation. The dataset layout includes canonical runnable benchmark rows, image pairs used by the slice, and public provenance files when available. The dataset is intended for benchmark and demo material, cross-model comparison on a shared public slice basket, and public reproducibility for a narrow part of the evaluation stack.