APT-36K

APT-36K的提供了现实场景中的动物姿态估计和跟踪标注。通过筛选TouTube视频，保留了了2400个有效视频片段，每个片段包含15帧，涵盖动物园、森林、沙漠等不同场景的30种不同动物。
发表于NeurIPS2023，https://github.com/pandorgan/APT-36K?tab=readme-ov-file#introduction

APT-36K provides annotations for animal pose estimation and tracking in real-world scenarios. By filtering YouTube videos, 2400 valid video clips are retained, each containing 15 frames. The dataset covers 30 distinct animal species across diverse settings including zoos, forests, and deserts. This work was published at NeurIPS 2023, with its official repository available at https://github.com/pandorgan/APT-36K?tab=readme-ov-file#introduction

搜集汇总

背景与挑战

背景概述

APT-36K是一个大规模动物姿态估计和跟踪数据集，包含2400个视频片段（每个15帧），涵盖30种动物和多种场景，由多个学术机构联合发布，数据集大小为21.6GB。