MVMO

MVMO数据集是由TECNALIA - 巴斯克研究和科技联盟创建的，包含116,000个场景，每个场景中有随机放置的10个不同类别的物体，从25个相机位置捕捉。数据集提供每个视角的语义分割地面实况，具有宽基线和物体高密度，适用于多视角语义分割和跨视角语义转移研究。创建过程中使用了Blender的Python API和ModelNet10数据集，每个场景包含15-20个物体，渲染为256x256像素的图像。MVMO数据集旨在解决单视角在严重遮挡场景中的局限性，推动多视角和跨视角语义分割技术的发展。

The MVMO dataset was developed by TECNALIA – Basque Research and Technology Alliance. It comprises 116,000 scenes, each containing 10 randomly placed objects from distinct categories, captured from 25 camera positions. The dataset provides semantic segmentation ground truth for each viewpoint, featuring wide baselines and high object density, making it suitable for research on multi-view semantic segmentation and cross-view semantic transfer. Constructed using Blender's Python API and the ModelNet10 dataset, each scene in the dataset includes 15–20 objects, with all scenes rendered into 256×256 pixel images. The MVMO dataset aims to address the limitations of single-view approaches in heavily occluded scenarios, and promote the advancement of multi-view and cross-view semantic segmentation technologies.

搜集汇总

背景与挑战

背景概述

MVMO数据集是一个多视角语义分割数据集，包含116,000个场景，每个场景随机放置10个不同类别的物体，从25个相机位置捕捉，提供语义分割地面实况。它具有宽基线和物体高密度的特点，使用Blender和ModelNet10创建，图像分辨率为256x256像素，旨在解决严重遮挡场景中单视角的局限性，适用于推动多视角和跨视角语义分割技术的研究。