欧亚猞猁个体识别数据集

1. 简介CzechLynx 数据集围绕欧亚猞猁的计算机视觉研究构建，包含约 37k 张手动验证和标记的图像，涵盖真实相机陷阱照片和合成样本，数据跨越 15 年以上，覆盖中欧两个地区（波西米亚西南部、喀尔巴阡山脉）；数据集按个体识别、姿势估计、实例分割三个任务划分子集（子集高度重叠，纳入标准和注释细节不同），提供多个子数据集及训练 / 验证 / 测试拆分；元数据丰富，包含身份标签、时间戳、位置等结构化信息，文件夹按源分类，分真实数据和合成数据目录；仅裁剪动物的重新识别子集公开，姿势估计和实例分割子集需签署保密协议获取1. Introduction The CzechLynx dataset is constructed for computer vision research focused on Eurasian lynx. It comprises approximately 37,000 manually verified and annotated images, including both real camera trap photographs and synthetic samples. The dataset spans over 15 years and covers two regions in Central Europe: southwestern Bohemia and the Carpathian Mountains. It is divided into subsets for three tasks: individual identification, pose estimation, and instance segmentation. The subsets have high overlap but differ in inclusion criteria and annotation details; multiple sub-datasets and train/validation/test splits are provided. The dataset features rich metadata including structured information such as identity labels, timestamps, and geographic locations. The folders are categorized by data source, with separate directories for real and synthetic data. Only the animal-cropped individual re-identification subset is publicly available, while the pose estimation and instance segmentation subsets require signing a non-disclosure agreement (NDA) for access.

搜集汇总

背景与挑战

背景概述

该数据集是一个专注于欧亚猞猁的计算机视觉数据集，包含约3.7万张手动验证和标记的图像，涵盖真实相机陷阱照片和合成样本，数据跨越15年以上，覆盖中欧两个地区。数据集按个体识别、姿势估计、实例分割三个任务划分子集，提供丰富的元数据如身份标签和时间戳，适用于野生动物监测相关的模型训练和评估，但仅重新识别子集公开，其他子集需保密协议获取。