XinhuiLiu001/M3ISR

M³ISR是一个用于3D场景重建和新视角合成的合成数据集，涵盖室内和室外环境的多视角、多模态捕捉。数据集包含5个类别（卧室、厨房、客厅、室外1和室外2），每个类别有5个不同的场景实例，共25个场景。每个场景有两种类型（动态场景和静态场景），并提供六个同步相机视角。数据集以1920×1080分辨率渲染，包含分割图、深度图和语义标注。采用帧级训练/测试分割方案，训练帧与测试帧比例为5:1。数据集支持3D重建、新视角合成、动态场景建模和多模态学习等任务。

M³ISR is a synthetic dataset for 3D scene reconstruction and novel view synthesis, featuring multi-view, multi-modal captures across indoor and outdoor environments. The dataset includes 5 categories (Bedroom, Kitchen, LivingRoom, Outdoor1, and Outdoor2) with 5 distinct scene instances each, totaling 25 scenes. Each scene has two types (MovingCameraDynamicScene and MovingCameraStaticScene) and six synchronized camera views. Rendered in a photo-realistic style at 1920×1080 resolution, the dataset provides segmentation maps, depth maps, and semantic annotations. A frame-level train/test split is applied with a 5:1 ratio. The dataset supports tasks such as 3D reconstruction, novel view synthesis, dynamic scene modeling, and multi-modal learning.