zhoubingyu/BaboonLand
收藏Hugging Face2026-04-01 更新2026-04-12 收录
下载链接：
https://hf-mirror.com/datasets/zhoubingyu/BaboonLand
下载链接
链接失效反馈官方服务：
资源简介：
---
license: cc0-1.0
task_categories:
- video-classification
- object-detection
tags:
- baboons
- video
- animal behavior
- behavior recognition
- annotation
- annotated video
- conservation
- drone
- UAV
- imbalanced
- Kenya
- Mpala Research Centre
pretty_name: >-
BaboonLand Dataset: Tracking Primates in the Wild and Automating Behaviour
Recognition from Drone Videos
description: >-
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.
size_categories:
- 1M<n<10M
language:
- en
configs:
- config_name: default
data_files:
- split: train
path: "Dataset Viewer/train.csv"
- split: validation
path: "Dataset Viewer/val.csv"
---
# Dataset Card for BaboonLand Dataset: Tracking Primates in the Wild and Automating Behaviour Recognition from Drone Videos
## Dataset Description
- **Homepage:** [BaboonLand Site](https://baboonland.xyz/)
- **Paper:** [BaboonLand Dataset: Tracking Primates in the Wild and Automating Behaviour Recognition from Drone Videos](https://link.springer.com/article/10.1007/s11263-025-02493-5)
- **arXiv:** [https://arxiv.org/pdf/2405.17698](https://arxiv.org/pdf/2405.17698)
### Dataset Summary
BaboonLand is an aerial drone video dataset of wild olive baboons (*Papio anubis*) collected over 21 consecutive days in Laikipia ([Mpala Research Centre](https://mpala.org)), Kenya, following three troops during morning and evening movements to and from sleeping sites. The dataset contains UAV footage across diverse environments (e.g., sleeping tree, river, rock, open savannah, cliff), with up to ~70 individuals per frame, yielding dense multi-object scenes from an overhead viewpoint.
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.
### Download & Reconstruct
BaboonLand is stored as split ZIP archives (`*.zip.part.*`) tracked with **Git LFS**. You can either download everything at once, or pull only a specific subset (Charades / Dataset / Tracking), then **concatenate parts**.
**Integrity check:** Compare the printed `md5sum` values with the reference hashes in `BaboonLand/manifest.json`.
---
### Option A — Download everything (all parts)
```bash
git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
cat BaboonLand/charades.zip.part* > BaboonLand/charades.zip
cat BaboonLand/dataset.zip.part* > BaboonLand/dataset.zip
cat BaboonLand/tracking.zip.part* > BaboonLand/tracking.zip
md5sum BaboonLand/charades.zip
md5sum BaboonLand/dataset.zip
md5sum BaboonLand/tracking.zip
rm -rf BaboonLand/charades.zip.part*
rm -rf BaboonLand/dataset.zip.part*
rm -rf BaboonLand/tracking.zip.part*
```
---
### Option B — Download only Charades part
```bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
git lfs pull --include="BaboonLand/charades.zip.part*"
cat BaboonLand/charades.zip.part* > BaboonLand/charades.zip
md5sum BaboonLand/charades.zip
rm -rf BaboonLand/charades.zip.part*
```
---
### Option C — Download only Dataset part
```bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
git lfs pull --include="BaboonLand/dataset.zip.part*"
cat BaboonLand/dataset.zip.part* > BaboonLand/dataset.zip
md5sum BaboonLand/dataset.zip
rm -rf BaboonLand/dataset.zip.part*
```
---
### Option D — Download only Tracking part
```bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
git lfs pull --include="BaboonLand/tracking.zip.part*"
cat BaboonLand/tracking.zip.part* > BaboonLand/tracking.zip
md5sum BaboonLand/tracking.zip
rm -rf BaboonLand/tracking.zip.part*
```
---
### Option E — Download only CVAT templates
```bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
git lfs pull --include="BaboonLand/cvat_templates"
```
---
### Directory Layout
```text
BaboonLand
/charades -> The dataset converted to Charades format to train and evaluate behavior
recognition models. You can download the generated dataset from our webpage
or you can generate it yourself.
...
/cvat_templates -> Templates to backup projects in CVAT and explore/adjust annotations.
/behavior.zip
/tracking.zip
/dataset
/video_1
/actions
/0.xml
/1.xml -> Behavior annotations for individual with ID=1
...
/n.xml
/mini-scenes -> Mini-scenes generated from video.mp4 and tracks.xml
/0.mp4
/1.mp4
...
/n.mp4
/timeline.jpg
/tracks.xml -> Tracks + bounding boxes (CVAT for video 1.1). Each track has a unique ID.
/video.mp4 -> Original drone video
/video_2
...
/video_n
...
/scripts
/requirements.txt
/tracks2mini-scenes.py
/dataset2charades.py
/charades2video.py
/charades2visual.py
/dataset2tracking.py
/tracking2ultralytics.py
/ultralytics2pyramid.py
/tracking -> Tracking split + (optionally) Ultralytics-format detection data.
...
/README.md
```
### Supported Tasks and Leaderboards
#### Detection
We evaluate YOLOv8-X model with input resolution of 768x768 on our dataset and report mAP@50, Precision, and Recall:
| Model | mAP@50 | Precision | Recall |
| --- | ---: | ---: | ---: |
| YOLOv8-X | 92.62 | 93.70 | 87.60 |
#### Tracking
We evaluate SORT, DeepSORT, StrongSORT, ByteTrack, and BotSort tracking algorithms on our dataset and report MOTA, MOTP, IDF1, Precision, and Recall:
| Tracker | MOTA | MOTP | IDF1 | Precision | Recall |
| --- | ---: | ---: | ---: | ---: | ---: |
| SORT | 84.76 | 50.15 | 77.43 | 90.83 | 91.19 |
| DeepSORT | 84.40 | 87.22 | 81.38 | 90.26 | 91.57 |
| StrongSORT | 82.48 | 85.37 | 84.98 | 88.00 | 90.10 |
| ByteTrack | 63.55 | 34.10 | 77.01 | 96.32 | 64.90 |
| BotSort | 63.81 | 34.31 | 78.24 | 97.21 | 66.16 |
#### Behavior Classes:
- Walking/Running
- Sitting/Standing
- Fighting/Playing
- Self-Grooming
- Being Groomed
- Grooming Somebody
- Mutual Grooming
- Infant-Carrying
- Foraging
- Drinking
- Mounting
- Sleeping
- Occluded
#### Behavior Recognition
We evaluate I3D, SlowFast, and X3D models on our dataset and report Micro-Average (Per Instance) and Macro-Average (Per Class) accuracy.
| Method | WI | Micro Top-1 | Micro Top-3 | Micro Top-5 | Macro Top-1 | Macro Top-3 | Macro Top-5 |
| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: |
| I3D | Random | 61.29 | 89.38 | 92.34 | 26.53 | 54.51 | 65.47 |
| SlowFast | Random | 61.71 | 90.35 | 93.11 | 27.08 | 56.73 | 67.61 |
| X3D | Random | 63.97 | 91.34 | 95.17 | 30.04 | 60.58 | 72.13 |
| X3D | K-400 | 64.89 | 92.54 | 96.66 | 31.41 | 62.04 | 74.01 |
### Languages
English
## Dataset Structure
BaboonLand provides original videos, CVAT-formatted annotations, derived mini-scenes, and scripts to generate task-specific training formats (e.g., Ultralytics/YOLO and Charades for SlowFast).
### Data Instances
Each `dataset/video_k/` directory contains:
- `video.mp4`: original UAV video
- `tracks.xml`: per-frame tracks (IDs + bounding boxes)
- `actions/*.xml`: per-track behavior labels (filename matches track ID)
- `mini-scenes/*.mp4`: cropped clips centered on each tracked individual (filename matches track ID)
### Data Fields
BaboonLand supports three derived tasks:
- **Detection:** bounding boxes for baboons (also convertible to Ultralytics/YOLO format via provided scripts).
- **Tracking:** per-frame tracks with persistent IDs and bounding boxes (stored in simplified CVAT for video 1.1).
- **Behavior recognition:** per-individual **mini-scenes** (cropped clips centered on each tracked individual) labeled into **12 behavior classes + Occluded**.
### Data Splits
BaboonLand includes task-specific evaluation sets:
- **Tracking:** 75% of each video for training, 25% for testing.
- **Detection (YOLO-formatted):** 80% training, 7% validation, 13% testing.
- **Behavior recognition (Charades format):** 75% training, 25% testing.
#### Data Collection and Procedures
- **Species:** Olive baboons (*Papio anubis*)
- **Location:** Mpala Research Centre, Laikipia County, Kenya
- **Capture:** DJI Air 2S, videos recorded in **5.3K**
- **Procedure:** all flights were conducted above 20 meters from a closes animal.
## Personal and Sensitive Information
- No humans can be distinguished in the videos.
- Data collection followed research licensing and animal care protocols (see Acknowledgments).
### Authors
- Isla Duporge
- Maksim Kholiavchenko
- Roi Harel
- Scott Wolf
- Daniel Rubenstein
- Meg Crofoot
- Tanya Berger-Wolf
- Stephen Lee
- Julie Barreau
- Jenna Kline
- Michelle Ramirez
- Charles Stewart
### Citation Information
#### Dataset
####
```
@misc{duporge2026baboonland_hf,
title = {BaboonLand Dataset: Tracking Primates in the Wild and Automating Behaviour Recognition from Drone Videos},
author = {Isla Duporge and Maksim Kholiavchenko and Roi Harel and Scott Wolf and Daniel Rubenstein and Margaret Crofoot and Tanya Berger-Wolf and Stephen Lee and Julie Barreau and Jenna Kline and Michelle Ramirez and Charles Stewart},
year = 2026,
url = {https://huggingface.co/datasets/imageomics/BaboonLand},
doi = {10.57967/hf/7470},
publisher = {Hugging Face}
}
```
#### Paper
```
@article{duporge2025baboonland,
title={BaboonLand Dataset: Tracking Primates in the Wild and Automating Behaviour Recognition from Drone Videos},
author={Duporge, Isla and Kholiavchenko, Maksim and Harel, Roi and Wolf, Scott and Rubenstein, Daniel I and Crofoot, Margaret C and Berger-Wolf, Tanya and Lee, Stephen J and Barreau, Julie and Kline, Jenna and Ramirez, Michelle and Stewart, Charles},
journal={International Journal of Computer Vision},
pages={1--12},
year={2025},
publisher={Springer}
}
```
### Contributions / Acknowledgments
This material is based upon work supported by the US National Science Foundation (NSF) through the [Imageomics Institute](https://imageomics.org) (NSF HDR program) under [Award No. 2118240](https://www.nsf.gov/awardsearch/showAward?AWD_ID=2118240) (“Imageomics: A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning”) and through the [AI Institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE)](https://icicle.osu.edu/) under [Award No. 2112606](https://www.nsf.gov/awardsearch/showAward?AWD_ID=2112606). ID was supported by the National Academy of Sciences Research Associate Program and the United States Army Research Laboratory while conducting this study. ID collected all the UAV data on a Civil Aviation Authority Drone License CAA NQE Approval Number: 0216/1365 in conjunction with authorization from a KCAA operator under a Remote Pilot License. The data were gathered at the [Mpala Research Centre](https://mpala.org/) in Kenya, in accordance with Research License No. NACOSTI/P/22/18214. The data collection protocol adhered strictly to the guidelines set forth by the Institutional Animal Care and Use Committee under permission No. IACUC 1835F. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
许可证：CC0 1.0协议
任务类别：
- 视频分类
- 目标检测
标签：
- 狒狒
- 视频
- 动物行为
- 行为识别
- 标注
- 带标注视频
- 物种保护
- 无人机（Drone）
- 无人机（UAV）
- 样本不均衡
- 肯尼亚
- 姆帕拉研究中心（Mpala Research Centre）
友好名称：BaboonLand数据集：野外灵长类追踪与无人机视频行为识别自动化
数据集描述：BaboonLand是针对肯尼亚莱基皮亚地区野生橄榄狒狒构建的航拍无人机视频数据集，采集时长跨越连续21天，覆盖三个狒狒群体。该数据集包含20余小时的影像素材，画面中存在高密度多个体场景（单帧最多可容纳约70只狒狒），并附带支持目标检测、多目标追踪与行为识别任务的标注信息。
数据规模类别：1M<n<10M
语言：英语
配置项：
- 配置名称：default
数据文件：
- 拆分集：训练集
路径："Dataset Viewer/train.csv"
- 拆分集：验证集
路径："Dataset Viewer/val.csv"
# 狒狒乐园（BaboonLand）数据集卡片：野外灵长类追踪与无人机视频行为识别自动化
## 数据集概述
- 【主页】：[狒狒乐园官网](https://baboonland.xyz/)
- 【论文】：[《BaboonLand数据集：野外灵长类追踪与无人机视频行为识别自动化》](https://link.springer.com/article/10.1007/s11263-025-02493-5)
- 【arXiv预印本】：[https://arxiv.org/pdf/2405.17698](https://arxiv.org/pdf/2405.17698)
### 数据集详情
狒狒乐园数据集是针对肯尼亚莱基皮亚地区（[姆帕拉研究中心](https://mpala.org)）野生橄榄狒狒（*Papio anubis*）构建的航拍视频数据集，采集周期为连续21天，追踪了三个狒狒群体在每日往返睡眠位点的晨间与晚间活动。该数据集包含无人机（UAV）拍摄的多场景影像，涵盖休眠树、河流、岩石、开阔稀树草原、悬崖等多种环境，单帧最多可容纳约70只个体，从鸟瞰视角呈现高密度多目标场景。
该数据集支持三大核心子任务：目标检测、多目标追踪与行为识别。其包含三部分内容：(1) 基于约5.3K分辨率帧通过多尺度分块生成的检测数据集（约3万张图像）；(2) 约0.5小时的高密度追踪标注；(3) 约20小时的行为“微场景”标注，这些微场景被划分为12种行为类别，以及一个遮挡额外类别。
### 下载与重建
狒狒乐园数据集以分卷压缩包（`*.zip.part.*`）形式存储，通过**Git大文件存储（Git Large File Storage, Git LFS）**进行管理。用户可选择一次性下载全部数据，或仅拉取指定子集（Charades/数据集/追踪集），随后合并分卷文件。
【完整性校验】：将生成的`md5sum`校验值与`BaboonLand/manifest.json`中的参考哈希值进行比对。
---
### 选项A — 下载全部数据
bash
git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
cat BaboonLand/charades.zip.part* > BaboonLand/charades.zip
cat BaboonLand/dataset.zip.part* > BaboonLand/dataset.zip
cat BaboonLand/tracking.zip.part* > BaboonLand/tracking.zip
md5sum BaboonLand/charades.zip
md5sum BaboonLand/dataset.zip
md5sum BaboonLand/tracking.zip
rm -rf BaboonLand/charades.zip.part*
rm -rf BaboonLand/dataset.zip.part*
rm -rf BaboonLand/tracking.zip.part*
---
### 选项B — 仅下载Charades子集
bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
git lfs pull --include="BaboonLand/charades.zip.part*"
cat BaboonLand/charades.zip.part* > BaboonLand/charades.zip
md5sum BaboonLand/charades.zip
rm -rf BaboonLand/charades.zip.part*
---
### 选项C — 仅下载数据集子集
bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
git lfs pull --include="BaboonLand/dataset.zip.part*"
cat BaboonLand/dataset.zip.part* > BaboonLand/dataset.zip
md5sum BaboonLand/dataset.zip
rm -rf BaboonLand/dataset.zip.part*
---
### 选项D — 仅下载追踪集子集
bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
git lfs pull --include="BaboonLand/tracking.zip.part*"
cat BaboonLand/tracking.zip.part* > BaboonLand/tracking.zip
md5sum BaboonLand/tracking.zip
rm -rf BaboonLand/tracking.zip.part*
---
### 选项E — 仅下载CVAT模板
bash
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/imageomics/BaboonLand
cd BaboonLand
git lfs pull --include="BaboonLand/cvat_templates"
---
### 目录结构
text
BaboonLand
/charades -> 转换为Charades格式的数据集，用于训练和评估行为识别模型。可从官网下载生成后的数据集，或自行生成。
...
/cvat_templates -> 用于备份CVAT项目、探索或调整标注的模板文件
/behavior.zip
/tracking.zip
/dataset
/video_1
/actions
/0.xml
/1.xml -> 编号为1的个体的行为标注
...
/n.xml
/mini-scenes -> 从video.mp4和tracks.xml生成的微场景片段
/0.mp4
/1.mp4
...
/n.mp4
/timeline.jpg
/tracks.xml -> 追踪序列与边界框（CVAT视频1.1简化格式），每个追踪序列拥有唯一ID
/video.mp4 -> 原始无人机航拍视频
/video_2
...
/video_n
...
/scripts
/requirements.txt
/tracks2mini-scenes.py
/dataset2charades.py
/charades2video.py
/charades2visual.py
/dataset2tracking.py
/tracking2ultralytics.py
/ultralytics2pyramid.py
/tracking -> 追踪集拆分+（可选）Ultralytics格式的检测数据
...
/README.md
### 支持任务与排行榜
#### 目标检测
我们使用输入分辨率为768x768的YOLOv8-X模型在本数据集上进行评估，报告mAP@50、精确率与召回率指标：
| 模型 | mAP@50 | 精确率 | 召回率 |
| --- | ---: | ---: | ---: |
| YOLOv8-X | 92.62 | 93.70 | 87.60 |
#### 多目标追踪
我们在本数据集上评估了SORT、DeepSORT、StrongSORT、ByteTrack与BotSort五种追踪算法，报告MOTA、MOTP、IDF1、精确率与召回率指标：
| 追踪算法 | MOTA | MOTP | IDF1 | 精确率 | 召回率 |
| --- | ---: | ---: | ---: | ---: | ---: |
| SORT | 84.76 | 50.15 | 77.43 | 90.83 | 91.19 |
| DeepSORT | 84.40 | 87.22 | 81.38 | 90.26 | 91.57 |
| StrongSORT | 82.48 | 85.37 | 84.98 | 88.00 | 90.10 |
| ByteTrack | 63.55 | 34.10 | 77.01 | 96.32 | 64.90 |
| BotSort | 63.81 | 34.31 | 78.24 | 97.21 | 66.16 |
#### 行为类别：
- 行走/奔跑
- 坐/站立
- 打斗/玩耍
- 自我理毛
- 接受理毛
- 为其他个体理毛
- 互相理毛
- 携带幼崽
- 觅食
- 饮水
- 爬跨
- 睡眠
- 遮挡
#### 行为识别
我们在本数据集上评估了I3D、SlowFast与X3D三种模型，报告微平均（按实例统计）与宏平均（按类别统计）准确率：
| 方法 | 初始化方式 | Micro Top-1 | Micro Top-3 | Micro Top-5 | Macro Top-1 | Macro Top-3 | Macro Top-5 |
| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: |
| I3D | 随机初始化 | 61.29 | 89.38 | 92.34 | 26.53 | 54.51 | 65.47 |
| SlowFast | 随机初始化 | 61.71 | 90.35 | 93.11 | 27.08 | 56.73 | 67.61 |
| X3D | 随机初始化 | 63.97 | 91.34 | 95.17 | 30.04 | 60.58 | 72.13 |
| X3D | K-400预训练 | 64.89 | 92.54 | 96.66 | 31.41 | 62.04 | 74.01 |
### 语言
英语
## 数据集结构
狒狒乐园数据集提供原始视频、CVAT格式标注、衍生微场景，以及用于生成特定任务训练格式的配套脚本（例如Ultralytics/YOLO格式与用于SlowFast的Charades格式）。
### 数据实例
每个`dataset/video_k/`目录包含以下文件：
- `video.mp4`：原始无人机（UAV）航拍视频
- `tracks.xml`：逐帧追踪信息（包含唯一ID与边界框）
- `actions/*.xml`：单追踪序列的行为标签（文件名与追踪ID对应）
- `mini-scenes/*.mp4`：以每个追踪个体为中心的裁剪片段（文件名与追踪ID对应）
### 数据字段
狒狒乐园数据集支持三类衍生任务：
- **目标检测**：狒狒的边界框标注（可通过配套脚本转换为Ultralytics/YOLO格式）。
- **多目标追踪**：带持久化ID与边界框的逐帧追踪信息（以简化版CVAT视频1.1格式存储）。
- **行为识别**：以每个追踪个体为中心的裁剪微场景，标注为12种行为类别+遮挡类别。
### 数据拆分
狒狒乐园数据集包含针对特定任务的评估拆分：
- **追踪任务**：每个视频的75%用于训练，25%用于测试。
- **目标检测（YOLO格式）**：80%训练集、7%验证集、13%测试集。
- **行为识别（Charades格式）**：75%训练集、25%测试集。
#### 数据采集流程
- **物种**：橄榄狒狒（*Papio anubis*）
- **采集地点**：肯尼亚莱基皮亚县姆帕拉研究中心
- **采集设备**：DJI Air 2S，视频分辨率为**5.3K**
- **采集规范**：所有飞行均在距离最近动物20米以上的空域进行。
## 个人与敏感信息
- 视频中无法识别到任何人类个体。
- 数据采集遵循研究许可与动物护理规范（详见致谢部分）。
### 作者
- Isla Duporge（伊斯拉·杜波奇）
- Maksim Kholiavchenko（马克西姆·霍利亚夫琴科）
- Roi Harel（罗伊·哈雷尔）
- Scott Wolf（斯科特·沃尔夫）
- Daniel Rubenstein（丹尼尔·鲁宾斯坦）
- Meg Crofoot（梅格·克罗夫特）
- Tanya Berger-Wolf（坦尼娅·伯杰-沃尔夫）
- Stephen Lee（斯蒂芬·李）
- Julie Barreau（朱莉·巴罗）
- Jenna Kline（珍娜·克莱因）
- Michelle Ramirez（米歇尔·拉米雷斯）
- Charles Stewart（查尔斯·斯图尔特）
### 引用信息
#### 数据集引用
bibtex
@misc{duporge2026baboonland_hf,
title = {BaboonLand数据集：野外灵长类追踪与无人机视频行为识别自动化},
author = {Isla Duporge and Maksim Kholiavchenko and Roi Harel and Scott Wolf and Daniel Rubenstein and Margaret Crofoot and Tanya Berger-Wolf and Stephen Lee and Julie Barreau and Jenna Kline and Michelle Ramirez and Charles Stewart},
year = 2026,
url = {https://huggingface.co/datasets/imageomics/BaboonLand},
doi = {10.57967/hf/7470},
publisher = {Hugging Face}
}
#### 论文引用
bibtex
@article{duporge2025baboonland,
title={BaboonLand数据集：野外灵长类追踪与无人机视频行为识别自动化},
author={Duporge, Isla and Kholiavchenko, Maksim and Harel, Roi and Wolf, Scott and Rubenstein, Daniel I and Crofoot, Margaret C and Berger-Wolf, Tanya and Lee, Stephen J and Barreau, Julie and Kline, Jenna and Ramirez, Michelle and Stewart, Charles},
journal={国际计算机视觉杂志（International Journal of Computer Vision）},
pages={1--12},
year={2025},
publisher={Springer}
}
### 贡献与致谢
本研究工作受美国国家科学基金会（National Science Foundation, NSF）资助，通过[Imageomics研究所](https://imageomics.org)（NSF HDR项目，编号Award No. 2118240，项目名称“Imageomics：知识引导机器学习驱动的生物信息学新前沿”），以及[智能网络基础设施与计算学习环境人工智能研究所（ICICLE）](https://icicle.osu.edu/)（编号Award No. 2112606）提供支持。ID受美国国家科学院研究助理计划与美国陆军研究实验室资助开展本研究。ID持有民用航空管理局无人机执照（CAA NQE批准号：0216/1365），并在KCAA运营商授权下使用远程飞行员执照完成所有航拍数据采集。数据采集于肯尼亚姆帕拉研究中心（[Mpala Research Centre](https://mpala.org/)），遵循研究许可编号NACOSTI/P/22/18214。数据采集方案严格遵循机构动物护理与使用委员会指南，许可编号IACUC 1835F。本材料中的任何观点、发现、结论或建议仅代表作者本人，不一定反映美国国家科学基金会的官方立场。
提供机构：
zhoubingyu
