Amar-S/MOVi-MC-AC

MOVi-MC-AC数据集是一个包含多对象视频、多摄像头和模态内容的开放数据集。它是第一个包含被遮挡物体的模态内容真实标注的数据集，拥有约580万个实例，是迄今为止最大的模态数据集。该数据集由通用3D物体组成，这些物体在多摄像头视频中相互碰撞、弹跳。所有数据都是使用开源数据集生成器Kubric创建的，特别是我们修改了MOVi挑战，以包括额外的（6）摄像头和物体的未遮挡视图。数据集中的3D物体来自Google的Kubric/MOVi引擎默认物体，这些物体来自公开的Google扫描物体数据集。MOVi-MC-AC数据集支持多种计算机视觉任务，包括图像分割、视频对象分割、对象检测与分类、对象跟踪、跨视图对象重识别（多视图）和跨视频对象重识别。

The MOVi-MC-AC dataset is an open dataset for multi-object video with multiple cameras and amodal content. It is the first dataset to include ground-truth annotations for amodal content of obscured objects, with approximately 5.8 million instances, making it the largest amodal dataset to date. The dataset consists of video scenes of generic 3D objects thrown together, colliding with and bouncing off each other, captured from multiple camera perspectives. All data is created using the open-source dataset generator Kubric, with modifications to include additional cameras and unoccluded views of objects. The 3D objects used are from Googles Kubric/MOVi engine, sourced from the publicly available Google Scanned Objects dataset. MOVi-MC-AC supports various computer vision tasks, including image segmentation, video object segmentation, object detection, tracking, and re-identification across views and videos. The dataset also includes modal and amodal object masks, RGB content, and depth masks, providing a rich resource for research and development in computer vision.