Matched Frames Subset of the BAMBI UAV Wildlife Dataset

This dataset is part of the BAMBI (Biodiversity Airborne Monitoring Based on Intelligent UAV sampling; https://www.bambi.eco/) project. It contains multimodal nadir UAV recordings of forest wildlife captured with thermal and RGB sensors. This sub dataset provides a cross-modal evaluation subset derived from the BAMBI UAV wildlife dataset and its initial, iterative versions. It contains 5,305 matched RGB–thermal labelled image pairs with 13,057 co-registered bounding box annotations, enabling direct comparison between both sensing modalities. The pairs were generated using a multi-stage template matching pipeline that spatially aligns detections between thermal and RGB frames despite sensor synchronization offsets. Each matched pair contains corresponding crops and bounding boxes representing the same detected animal instance in both modalities. The dataset supports research on multimodal wildlife detection, cross-modal perception, modality comparison, and RGB–thermal fusion methods in challenging forest environments captured from nadir UAV imagery. The subset is designed as a paired benchmark for evaluating RGB and thermal models under identical spatial conditions, facilitating reproducible experiments on cross-modal learning and modality-specific performance in wildlife monitoring tasks. To create this matched dataset (labels_thermal_matched and labels_rgb_matched), we iteratively created initial versions with labels_thermal_original first, which was refined to labels_thermal_merged later. Based on that the initial RGB dataset (labels_rgb) was created, which was used to destill the final matched versions. You can find models trained on this dataset on HugginFace: https://huggingface.co/cpraschl/bambi-models The files follow the naming convention <flight_id>_<frame_idx>.<suffix>, where the flight_id corresponds to the videos published as part of the original dataset. The original dataset is split in multiple parts and can be found here: 10.5281/zenodo.18692354 10.5281/zenodo.18698508 10.5281/zenodo.18703312 10.5281/zenodo.18705705 10.5281/zenodo.18707610 10.5281/zenodo.18711217 10.5281/zenodo.18715162 10.5281/zenodo.18717601 You can find scripts for working with the original dataset at https://github.com/bambi-eco/Dataset. e.g. downloading the corresponding flight videos and allowing to extract the specific frames.本数据集隶属于BAMBI（基于智能无人机采样的生物多样性航空监测，Biodiversity Airborne Monitoring Based on Intelligent UAV sampling；https://www.bambi.eco/）项目，包含利用热成像与RGB传感器采集的森林野生动物天底视角无人机多模态录制数据。 该子数据集源自BAMBI无人机野生动物数据集及其初始迭代版本，是专为跨模态评估构建的子集。其包含5305对匹配的RGB-热成像标注图像对，以及13057个共配准边界框标注，可直接对比两种传感模态的表现。该图像对通过多阶段模板匹配流水线生成，即便存在传感器同步偏移，也能实现热成像与RGB帧间检测结果的空间对齐。 每一对匹配图像均包含对应裁剪区域与边界框，用于表征两种模态下检测到的同一野生动物实例。本数据集支持针对天底视角无人机拍摄的复杂森林环境下的多模态野生动物检测、跨模态感知、模态对比以及RGB-热成像融合方法的相关研究。 该子集被设计为配对基准测试集，用于在相同空间条件下评估RGB与热成像模型，助力跨模态学习与野生动物监测任务中模态特异性性能的可复现实验。为构建该匹配数据集（labels_thermal_matched与labels_rgb_matched），我们首先迭代生成初始版本labels_thermal_original，后续将其优化为labels_thermal_merged；以此为基础创建初始RGB数据集labels_rgb，最终通过该数据集提炼得到最终匹配版本。 基于本数据集训练的模型可在HuggingFace平台获取：https://huggingface.co/cpraschl/bambi-models 数据集文件遵循<flight_id>_<frame_idx>.<suffix>命名规则，其中flight_id对应原始数据集中发布的视频编号。原始数据集分为多个部分，可通过以下链接获取： 10.5281/zenodo.18692354 10.5281/zenodo.18698508 10.5281/zenodo.18703312 10.5281/zenodo.18705705 10.5281/zenodo.18707610 10.5281/zenodo.18711217 10.5281/zenodo.18715162 10.5281/zenodo.18717601 相关处理脚本（如下载对应飞行视频、提取指定帧等）可在GitHub仓库https://github.com/bambi-eco/Dataset获取。