Lightning Pose dataset: CRIM13

The Caltech Resident-Intruder Mouse dataset (CRIM13) (Burgos-Artizzu et al., CVPR 2012) consists of two mice interacting in an enclosed arena, captured by top and side view cameras at 30 Hz. We only use the top view. Seven keypoints are labeled on each mouse for a total of 14 keypoints (Segalin et al., eLife 2021).Each keypoint in the original CRIM13 dataset (https://data.caltech.edu/records/4emt5-b0t10) is labeled by five different annotators. To create the final set of labels, we take the median across all labels for each keypoint. Additionally, we remove all frames where one or both mice were absent.The labeled data are partitioned into disjoint "in-distribution" (InD) and "out-of-distribution" (OOD) sets. Each set contains different sessions / animals. We use the train/test split provided in the original dataset - the (4) resident mice are present in both InD and OOD splits; however, the intruder mouse is different for each session. The InD data contain 3986 labeled frames, and 37 unlabeled videos; the OOD data contain 1274 labeled frames, and 19 unlabeled videos.Many thanks to the authors of the CRIM13 paper who collected and analyzed the original video dataset: Xavier P. Burgos-Artizzu, Piotr Dollár, Dayu Lin, David J. Anderson and Pietro Perona.We also thank the authors of the MARS paper who collected keypoint annotations for the CRIM13 dataset: Cristina Segalin, Jalani Williams, Tomomi Karigo, May Hui, Moriel Zelikowsky, Jennifer J Sun, Pietro Perona, David J Anderson and Ann Kennedy.加州理工学院驻地-入侵者小鼠数据集（CRIM13）（Burgos-Artizzu等，CVPR 2012）包含在封闭实验竞技场中互动的两只小鼠，由顶视与侧视摄像头以30Hz帧率采集，本研究仅采用顶视角数据。每只小鼠标注有7个关键点，总计14个关键点（Segalin等，eLife 2021）。 原始CRIM13数据集（https://data.caltech.edu/records/4emt5-b0t10）中的每个关键点均由5名不同标注人员完成标注。为生成最终标注集，我们对每个关键点的所有标注结果取中位数。此外，我们剔除了单只或两只小鼠缺失的所有帧。 标注数据被划分为互不重叠的分布内（in-distribution，简称InD）与分布外（out-of-distribution，简称OOD）两个子集，每个子集包含不同的实验会话与实验动物。我们采用原始数据集提供的训练/测试划分：4只驻地小鼠同时出现在InD与OOD划分中，但不同实验会话中的入侵者小鼠各不相同。InD子集包含3986张标注帧与37个未标注视频；OOD子集包含1274张标注帧与19个未标注视频。 衷心感谢CRIM13论文的作者们收集并分析了原始视频数据集：Xavier P. Burgos-Artizzu、Piotr Dollár、Dayu Lin、David J. Anderson及Pietro Perona。我们同时感谢为CRIM13数据集收集关键点标注的MARS论文作者们：Cristina Segalin、Jalani Williams、Tomomi Karigo、May Hui、Moriel Zelikowsky、Jennifer J Sun、Pietro Perona、David J Anderson及Ann Kennedy。