DAP-2M
收藏资源简介:
DAP-2M是由Insta360 Research联合多所高校构建的大规模全景深度数据集,包含200万条跨室内外场景的合成与真实数据。该数据集整合了Structured3D的室内标注数据、基于UE5的AirSim360模拟器生成的9万张室外合成全景,以及从互联网爬取的170万张真实全景图像和DiT360生成的20万张室内补充样本。通过三阶段伪标签优化流程,有效弥合了合成与真实数据间的领域差异。该数据集为全景深度估计提供了多样化的监督信号,支撑了在机器人导航、空间感知等领域的应用,旨在解决现有全景数据规模不足和领域泛化能力弱的核心问题。
DAP-2M is a large-scale panoramic depth dataset developed by Insta360 Research in collaboration with multiple universities. It contains 2 million synthetic and real data samples spanning indoor and outdoor scenes. This dataset integrates indoor annotated data from Structured3D, 90,000 outdoor synthetic panoramic images generated by the UE5-based AirSim360 simulator, 1.7 million real panoramic images crawled from the Internet, and 200,000 supplementary indoor samples generated by DiT360. Through a three-stage pseudo-label optimization pipeline, it effectively narrows the domain gap between synthetic and real-world data. This dataset provides diverse supervision signals for panoramic depth estimation, supports applications in fields such as robot navigation and spatial perception, and aims to address the core challenges of insufficient scale of existing panoramic datasets and weak domain generalization capability.

- 1Depth Any Panoramas: A Foundation Model for Panoramic Depth EstimationInsta360 Research, 加州大学圣地亚哥分校, 武汉大学, 加州大学默塞德分校 · 2025年



