InteriorNet

系统概述: 用于渲染rgb-d惯性基准的端到端管道，用于大规模内部场景理解和映射。我们的数据集包含由pipeline创建的20M图像 :( A) 我们收集了世界领先家具制造商提供的大约100万个CAD模型。这些模型已在现实世界的生产中使用。(B) 基于这些模型，大约1,100名专业设计师围绕2200万室内布局创建。大多数此类布局已用于现实世界的装饰中。(C) 对于每种布局，我们都会生成许多配置，以表示不同的随机照明和模拟日常生活中随时间变化的场景。(D) 我们提供了一个交互式模拟器 (ViSim)，以帮助创建地面真相IMU，事件以及单目或立体相机轨迹，包括手绘，随机行走和基于神经网络的逼真轨迹。(E) 所有支持的图像序列和地面真相。

System Overview: An end-to-end pipeline for rendering RGB-D inertial benchmarks for large-scale indoor scene understanding and mapping. Our dataset contains 20 million images created by this pipeline: (A) We collected approximately 1 million CAD models provided by the world's leading furniture manufacturers. These models have been used in real-world production. (B) Based on these models, approximately 1,100 professional designers created over 22 million indoor layouts. Most of these layouts have been applied in real-world interior decoration. (C) For each layout, we generate numerous configurations to represent different random lighting conditions and simulate scenarios that change over time in daily life. (D) We provide an interactive simulator (ViSim) to assist in creating ground-truth IMU data, event data, and monocular or stereo camera trajectories, including hand-drawn, random walk, and neural network-based realistic trajectories. (E) All supported image sequences and ground-truth data are provided.

搜集汇总

背景与挑战

背景概述

InteriorNet是一个大规模室内场景数据集，包含约2000万张图像，专为场景理解和映射任务设计。其特点在于基于真实CAD模型和设计师创建的布局，生成多样化的照明和动态场景配置，并提供交互式模拟器ViSim以支持地面真相数据的生成。该数据集由伦敦帝国理工学院于2018年发布，适用于非商业研究用途。