imageomics/BaboonLand

BaboonLand是一个关于野生橄榄狒狒的无人机航拍视频数据集，收集于肯尼亚Laikipia地区，连续21天跟踪三个群体。数据集包含超过20小时的视频素材，密集的多个体场景（每帧最多约70只狒狒），并附有注释，支持检测、多目标跟踪和行为识别。数据集支持三个核心子任务：检测、多目标跟踪和行为识别。检测数据集来自约5.3K分辨率的帧，通过多尺度平铺（≈30K图像）；约0.5小时的密集跟踪注释；约20小时的行为“迷你场景”注释为12个行为类别和额外的遮挡类别。数据集还提供了多种下载选项和目录结构，以及支持的任务和排行榜。

BaboonLand is an aerial drone video dataset of wild olive baboons in Laikipia, Kenya, collected over 21 consecutive days across three troops. It contains 20+ hours of footage with dense multi-individual scenes (up to ~70 baboons per frame) and annotations enabling detection, multi-object tracking, and behavior recognition. The dataset supports three core subtasks: detection, multi-object tracking, and behavior recognition. It includes (1) a detection dataset derived from ~5.3K-resolution frames via multi-scale tiling (≈30K images), (2) ~0.5 hours of dense tracking annotations, and (3) ~20 hours of behavior “mini-scenes” annotated into 12 behavior classes and additional category for occlusions. The dataset also provides various download options, directory structures, and supported tasks and leaderboards.