DA2: Depth Anything in Any Direction
收藏资源简介:
DA2数据集是由腾讯混元、加州大学圣地亚哥分校和香港科技大学联合创建的,旨在解决全景图像深度估计问题。该数据集包含了约607,000对全景RGB深度图像,是通过将大量的透视图像转换为全景图像生成的。数据集的创建过程包括透视到等距圆柱投影的转换以及全景图像的补全。DA2数据集的应用领域包括全景多视图重建、家居装饰和机器人模拟等,旨在提供高质量的三维深度信息,推动全景应用场景的发展。
The DA2 dataset was jointly developed by Tencent Hunyuan, University of California, San Diego, and Hong Kong University of Science and Technology, with the core goal of solving the problem of panoramic image depth estimation. It contains approximately 607,000 pairs of panoramic RGB-depth images, which are generated by converting a large volume of perspective images into panoramic formats. The construction of this dataset involves two key steps: perspective-to-equirectangular projection transformation and panoramic image inpainting. The DA2 dataset can be applied in multiple fields including panoramic multi-view reconstruction, home decoration, and robotic simulation, and it is designed to provide high-quality 3D depth information to boost the development of panoramic application scenarios.

- 1通过腾讯混元, 加州大学圣地亚哥分校, 香港科技大学 · 2025年



