Rung 0 Dataset for Roman Strong Lens Data Challenge

Data Challenge Overview

The Roman Space Telescope is expected to observe O(10^5) galaxy-galaxy strong gravitational lenses, providing high angular resolution images of galaxy-galaxy strong gravitational lenses that can be used to probe the nature of dark matter at sub-galactic scales (Daylan and Birrer 2023, Wedig et al. 2025). The Roman Data Challenge for Dark Matter Substructure with Galaxy-Galaxy Strong Gravitational Lenses provides realistic simulated Roman images of strong lenses with various dark matter substructure populations and challenges the community to test out substructure detection and characterization pipelines.

Dataset Description

The goal of this rung is to serve as a tutorial for participants. You will familiarize yourself with the systematics of the challenge (e.g. where and how to submit, how to set up mejiro and obtain/generate datasets, where to find relevant documentation) and for us as the organizers to make sure that the data challenge runs smoothly. In this rung, you will train a regression model to determine the Einstein radius of lenses.

Changelog

v2.0: Uses romanisim instead of GalSim

Units of pixels are now DN/s (Roman L2) instead of counts

Effective exposure time is slightly shorter (610 instead of 642 seconds)

Images are 73x73 pixels (8.03") instead of 91x91 pixels (10.01")

v1.2: Correction of typo in dataset metadata. Units of pixels are counts, not counts/sec

v1.1: Correction of minor typos in Jupyter notebook

v1.0: All systems have substructure added to mass model

v0.1: Initial version with 50/50 substructure/no substructure split

数据挑战赛概述

罗曼空间望远镜（Roman Space Telescope）预计将观测O(10^5)量级的星系-星系强引力透镜（galaxy-galaxy strong gravitational lens），并获取这类透镜的高角分辨率图像，这些图像可用于探究亚星系尺度下的暗物质性质（Daylan与Birrer 2023，Wedig等人2025）。基于星系-星系强引力透镜的暗物质亚结构罗曼数据挑战赛（Roman Data Challenge for Dark Matter Substructure with Galaxy-Galaxy Strong Gravitational Lenses）提供了包含各类暗物质亚结构种群的逼真模拟罗曼望远镜强引力透镜图像，面向学界开放以测试亚结构探测与特征化分析流程。

数据集说明

本阶段旨在为参与者提供入门教程。您将熟悉本次挑战赛的相关流程规范（例如提交渠道与方式、如何配置mejiro工具并获取/生成数据集、如何查阅相关文档），同时主办方也可借此确保挑战赛顺利开展。在本阶段中，您将训练一个回归模型以测定透镜的爱因斯坦半径（Einstein radius）。

更新日志

v2.0: 改用romanisim而非GalSim；像素单位现为DN/s（罗曼L2级数据），而非计数（counts）；有效曝光时间略有缩短（从642秒调整为610秒）；图像尺寸调整为73×73像素（对应角尺寸8.03"），而非此前的91×91像素（10.01"）。
v1.2: 修正了数据集元数据中的一处拼写错误，像素单位为计数（counts），而非计数每秒。
v1.1: 修正了Jupyter Notebook中的一处微小拼写错误。
v1.0: 所有系统的质量模型均已加入亚结构。
v0.1: 初始版本，亚结构与无亚结构的样本占比为50:50。