https://ganko.asia/tags/datasets/Recent content in Datasets on Ganko SpaceHugozh-cnThu, 27 Aug 2026 00:00:00 +0000https://ganko.asia/posts/cvpr26_benchmark/Thu, 27 Aug 2026 00:00:00 +0000https://ganko.asia/posts/cvpr26_benchmark/<h2 id="the-sa-fari-dataset-segment-anything-in-footage-of-animals-for-recognition-and-identification">The SA-FARI Dataset: Segment Anything in Footage of Animals for Recognition and Identification</h2> <p>这篇论文提出了 SA-FARI——目前最大规模的开源野生动物多动物跟踪（Multi-Animal Tracking, MAT）数据集。</p> <p>野生动物保护依赖自动视频分析，而多动物跟踪（MAT）是其中的基础任务——它支撑个体重识别（re-identification）和行为识别等下游应用。但领域内缺乏合适的基准：规模小、物种单一、时空多样性不足。</p> <p>数据集核心指标：</p> <ul> <li>11,609 个视频，2014–2024 年采集，覆盖 741 个地点、4 大洲；</li> <li>~46 小时密集标注视频，99 个物种类别；</li> <li>16,224 个 masklet 身份（单个个体动物的时空分割轨迹）和942,702 个边界框/分割掩码/物种标签。</li> </ul> <p><strong>masklet</strong>：由分割掩码（而非仅有边界框）构成的时序轨迹，是比普通 tracklet 信息量更丰富的标注形式；并且数据附带匿名化的相机陷阱位置。</p> <table> <thead> <tr> <th>Dataset</th> <th>Sites</th> <th>Species</th> <th>Videos</th> <th>Duration (h)</th> <th>Tracklets</th> <th>Segmentation</th> <th>Source</th> </tr> </thead> <tbody> <tr> <td>AnimalTrack [51]</td> <td>—</td> <td>10</td> <td>58</td> <td>0.23</td> <td>2k</td> <td>✕</td> <td>YouTube</td> </tr> <tr> <td>GMOT [2] †</td> <td>—</td> <td>4</td> <td>16</td> <td>0.03</td> <td>1k</td> <td>✕</td> <td>YouTube</td> </tr> <tr> <td>TAO [17] / BURST [1] †</td> <td>—</td> <td>51</td> <td>429</td> <td>3.60</td> <td>1k</td> <td>✕/√</td> <td>Varied</td> </tr> <tr> <td>BuckTales [32]</td> <td>1</td> <td>1</td> <td>12</td> <td>0.21</td> <td>0.7k</td> <td>✕</td> <td>UAV</td> </tr> <tr> <td>BaboonLand [18]</td> <td>1</td> <td>1</td> <td>18</td> <td>0.32</td> <td>1k</td> <td>✕</td> <td>UAV</td> </tr> <tr> <td>KABR [26]</td> <td>1</td> <td>3</td> <td>742</td> <td>10.00</td> <td>—</td> <td>✕</td> <td>UAV</td> </tr> <tr> <td>WildLive [16]</td> <td>1</td> <td>3</td> <td>22</td> <td>0.70</td> <td>0.3k</td> <td>√</td> <td>UAV</td> </tr> <tr> <td>CCR [4]</td> <td>3</td> <td>1</td> <td>27</td> <td>10.00</td> <td>12k</td> <td>✕</td> <td>Video Camera</td> </tr> <tr> <td>MammAlps [20]</td> <td>3</td> <td>5</td> <td>2,384</td> <td>14.50</td> <td>6k</td> <td>✕</td> <td>Camera Trap</td> </tr> <tr> <td>PanAf500 [8]</td> <td>12</td> <td>2</td> <td>500</td> <td>2.10</td> <td>1.5k</td> <td>✕</td> <td>Camera Trap</td> </tr> <tr> <td>SA-FARI (ours)</td> <td>741</td> <td>99</td> <td>11,609</td> <td>45.78</td> <td>16k</td> <td>√</td> <td>Camera Trap</td> </tr> </tbody> </table> <p>支持任务列表：</p>