3D-ZeF

3D-ZeF是一个公开可用的立体3D RGB数据集，用于多目标斑马鱼跟踪。斑马鱼作为研究神经障碍、药物成瘾等的模型生物越来越受欢迎。行为分析是此类研究的关键部分。然而，斑马鱼的视觉相似性、遮挡和随机运动使得稳健的3D跟踪成为一个具有挑战性的未解决问题。该数据集包含八个序列，持续时间在15-120秒之间，包含1-10条自由游动的斑马鱼。视频已用总共86,400个点和边界框进行了标注。此外，我们还提出了一个复杂度评分和一个用于3D斑马鱼跟踪的新型开源模块化基线系统。该系统的表现是根据两个检测器：一个朴素方法和一个基于Faster R-CNN的鱼头检测器来衡量的。系统的MOTA最高可达77.6%。数据集的创建过程涉及使用标准玻璃水族箱和GoPro相机进行录制，并通过手动同步确保时间一致性。数据集的应用领域包括神经科学和生物学研究，旨在通过自动化跟踪系统解决手动检查的主观性和局限性。

3D-ZeF is a publicly available stereo 3D RGB dataset for multi-target zebrafish tracking. Zebrafish have grown increasingly popular as model organisms for research on neurological disorders, drug addiction and other related fields. Behavioral analysis constitutes a critical component of such research. However, the visual similarity among zebrafish, occlusions and their erratic movements make robust 3D tracking a challenging and unsolved problem. This dataset comprises eight sequences with durations ranging from 15 to 120 seconds, containing 1 to 10 freely swimming zebrafish. All videos have been annotated with a total of 86,400 points and bounding boxes. Additionally, we propose a complexity scoring metric and a novel open-source modular baseline system for 3D zebrafish tracking. The performance of this system is evaluated using two detectors: a naive baseline method and a Faster R-CNN-based fish head detector. The system achieves a maximum MOTA of 77.6%. The dataset was created using a standard glass aquarium and GoPro cameras for recording, with manual synchronization applied to ensure temporal consistency. The dataset finds applications in neuroscience and biological research, aiming to address the subjectivity and limitations of manual inspection via automated tracking systems.