gorilla-watch/Gorilla-SPAC-Wild

Gorilla-SPAC-Wild是一个用于西部低地大猩猩个体再识别的大规模视频数据集，旨在通过自动化分析摄像机陷阱视频来非侵入性地跟踪濒危大猩猩种群。数据集包含从刚果共和国Odzala-Kokoua国家公园的摄像机陷阱镜头中提取的图像，涵盖108个个体大猩猩在多种自然环境和光照条件下的多次遭遇。每个大猩猩都有高质量的面部（≥50×50像素）和全身裁剪图像注释，支持全面的再识别分析。数据集还包括跨遭遇数据，模拟了真实世界中因环境变化导致的外观变化的再识别挑战。数据集严格按个体划分，适用于新个体频繁出现的开放集评估场景。数据提取和处理使用了YOLOv8-Nano和BoostTrack++，并由具有15年以上实地经验的灵长类研究人员提供真实标签。数据集分为训练（70%）、验证（15%）、测试（15%）和单次遭遇子集。每个样本包含大猩猩的面部裁剪图像、个体标识符、捕获日期、视频来源、帧号、摄像机陷阱标识符和图像文件路径。

Gorilla-SPAC-Wild is a comprehensive benchmark dataset for individual re-identification of Western Lowland Gorillas from camera trap footage in natural rainforest environments. This dataset addresses a critical bottleneck in conservation: automating the analysis of vast archives of camera trap video to track endangered gorilla populations non-invasively. Extracted from camera trap footage at Odzala-Kokoua National Park, Republic of Congo, it features 108 individual gorillas tracked across multiple encounters in natural, challenging lighting and environmental conditions. Each gorilla is annotated with both high-quality facial crops (≥50×50 pixels) and full-body crops, enabling comprehensive re-identification analysis. The dataset includes cross-encounter data, simulating real-world re-identification challenges where appearance changes due to varying environmental conditions, and follows an open-set evaluation with strict individual-based splits. Data extraction and processing utilized YOLOv8-Nano and BoostTrack++, with ground truth labels provided by primate researchers with 15+ years of field experience. The dataset is split into Train (70%), Validation (15%), Test (15%), and Single-Encounter subsets. Each sample includes the gorillas face crop image, individual identifier, capture date, source video, frame number, camera trap identifier, and image file path.