TSGaussian
收藏资源简介:
TSGaussian是一个结合语义约束和深度先验的新型框架,旨在从稀疏视角中重建具有复杂结构的目标。该数据集通过YOLOv9和SAM生成2D语义掩码,并利用单目深度估计作为先验,结合3D空间一致性正则化来优化3D高斯分布。数据集的创建过程包括从360°稀疏图像序列中提取目标掩码和深度图,并通过语义控制和剪枝策略减少3D高斯的冗余。该数据集主要应用于计算机视觉领域,特别是在稀疏视角下的特定目标重建任务,旨在提高3D场景重建的精度和效率。
TSGaussian is a novel framework that integrates semantic constraints and deep priors, designed to reconstruct targets with complex structures from sparse viewpoints. This dataset generates 2D semantic masks via YOLOv9 and SAM, leverages monocular depth estimation as the prior, and combines 3D spatial consistency regularization to optimize 3D Gaussian distributions. The dataset creation process involves extracting target masks and depth maps from 360° sparse image sequences, and reducing the redundancy of 3D Gaussians through semantic control and pruning strategies. This dataset is primarily applied in the field of computer vision, particularly for specific target reconstruction tasks from sparse viewpoints, with the goal of enhancing the accuracy and efficiency of 3D scene reconstruction.
TSGaussian 数据集概述
数据集名称
TSGaussian
数据集描述
TSGaussian 是一个用于语义和深度引导的目标特定高斯喷射(Gaussian Splatting)的数据集,适用于从稀疏视角进行处理。




