MASSTAR

MASSTAR是一个多模态大规模场景数据集，由中山大学人工智能学院等机构创建，旨在解决复杂场景的表面预测和完成问题。数据集包含超过一千个场景级3D网格模型，结合了真实世界数据和多模态信息，如RGB图像、描述文本和深度图像。创建过程中，利用高效工具链从原始3D数据中筛选高质量模型并生成多模态数据。该数据集适用于机器人应用，如高质量3D重建和自动驾驶，旨在提高算法在复杂环境中的性能和鲁棒性。

MASSTAR is a large-scale multimodal scene dataset developed by institutions including the School of Artificial Intelligence, Sun Yat-sen University, targeting surface prediction and completion tasks in complex scenarios. The dataset contains over 1,000 scene-level 3D mesh models, which integrates real-world data and multimodal information such as RGB images, descriptive texts and depth images. During its construction, an efficient toolchain was employed to screen high-quality models from raw 3D data and generate the corresponding multimodal data. This dataset is applicable to robotic applications including high-quality 3D reconstruction and autonomous driving, with the goal of enhancing the performance and robustness of algorithms in complex environments.