CalMS21 (Caltech Mouse Social Interactions)

Caltech Mouse Social Interactions (CalMS21) 数据集是来自行为神经科学的多智能体数据集。该数据集由社会互动的轨迹数据组成，这些数据是从标准的常驻入侵者分析中自由行为的老鼠的视频中记录下来的。 CalMS21 数据集是 2021 年多智能体行为挑战赛的一部分。为了帮助加速行为研究，CalMS21 数据集提供了一个基准来评估自动行为分类方法在三种设置中的性能：(1) 用于在所有带注释的大型行为数据集上进行训练通过单个注释器，（2）用于样式迁移以学习行为定义中注释器之间的差异，以及（3）在有限的训练数据下学习感兴趣的新行为。该数据集包含 600 万帧未标记的交互小鼠跟踪姿势，以及超过 100 万帧具有跟踪姿势和相应帧级行为注释的帧。数据集的挑战是能够使用标记和未标记的跟踪数据准确地对行为进行分类，以及能够泛化到新的注释器和行为。

Caltech Mouse Social Interactions (CalMS21) Dataset is a multi-agent dataset from the field of behavioral neuroscience. This dataset comprises trajectory data capturing social interactions, recorded from videos of freely behaving mice in the standard resident-intruder assay. The CalMS21 dataset was part of the 2021 Multi-Agent Behavior Challenge. To accelerate behavioral research, the CalMS21 dataset offers a benchmark for evaluating the performance of automated behavioral classification methods across three distinct settings: (1) training on large annotated behavioral datasets with annotations from a single annotator, (2) style transfer to learn inter-annotator discrepancies in behavioral definitions, and (3) learning novel behaviors of interest under limited training data. The dataset includes 6 million unlabeled frames of tracked poses of interacting mice, alongside over 1 million frames with tracked poses and corresponding frame-level behavioral annotations. The core challenges of this dataset are to accurately classify behaviors using both labeled and unlabeled tracked pose data, as well as to generalize to new annotators and novel behaviors.

搜集汇总

背景与挑战

背景概述

CalMS21是一个多智能体行为神经科学数据集，包含老鼠社交互动的轨迹数据和行为注释，用于评估自动行为分类方法在不同设置下的性能。数据集包含大量标记和未标记的跟踪姿势帧，旨在促进行为研究的自动分类和泛化能力。