Modified Swiss Dwellings: a Machine Learning-ready Dataset for Floor Plan Auto-Completion at Scale

<em>Modified Swiss Dwellings</em>The Modified Swiss Dwellings (MSD) dataset is a <strong>machine learning-ready dataset for floor plan auto-completion at scale</strong>. The MSD dataset is derived from the Swiss Dwellings database (v3.0.0). The MSD dataset (train split) contains <strong>4167 floor plans</strong> of <strong>single- as well as multi-unit building complexes across Switzerland</strong>, hence extending the building scale w.r.t. of other well know floor plan datasets like the RPLAN dataset. Since the MSD dataset will be part of a challenge @ ICCV in Paris, 2023, October 3, the <strong>test split is not yet made public</strong>. This will be added after the submission deadline of the challenge, which will be around mid September 2023.<br><em>Cleaning, filtering, and processing</em>All cleaning, filtering, and processing is done in Python. The Swiss Dwellings database is cleaned and filtered on residential building complexes that have a minimum room count (&gt;10) and have at least 2 "Zone 2" rooms (<em>e.g.</em>, living room, corridor, kitchen, dining). A graph extraction algorithm fully based on the `shapely` and `networkx` libraries in Python was developed to extract the access graphs from the filtered floor plans.<br><em>Dataset structure</em>The MSD dataset contains 3 files.<br>1) A README.md file explaining the dataset.2) A training set ZIP archive, containing 4 folders: `graph_in` [&lt;index&gt;.pickle], `struct_in` [&lt;index&gt;.npy], `full_out` [&lt;index&gt;.npy], and `graph_out` [&lt;index&gt;.pickle]. Naming is consistent across all folders, meaning that an instance from `graph_in` with name "&lt;index&gt;.pickle" is from the same floor plan as an instance from `full_out` with name "&lt;index&gt;.npy".3) A test set ZIP archive, containing 2 folders: `graph_in` and `struct_in` (similarly structured as the training data; but obviously with withheld annotations.)<br><em>Floor plan auto-completion</em>The MSD dataset is developed with the goal for the computer science community to develop (deep learning) models for the task of floor plan auto-completion. The floor plan auto-completion task takes as input the boundary of a building, the structural elements necessary for the building’s structural integrity, and a set of user constraints formalized in a graph structure, with the goal of automatically generating the full floor plan. Specifically, the goal is to learn the correlation between the the joint distribution of `graph_in` and `struct_in` with that of `full_out`. `graph_out` is provided when researchers want to use / develop methods from graph signal processing, or graph machine learning specifically.<br><em>GIthub guidelines</em>Github page.

## 改进版瑞士住宅数据集（Modified Swiss Dwellings，简称MSD）
改进版瑞士住宅数据集（Modified Swiss Dwellings，以下简称MSD）是一款专为大规模户型图自动补全任务打造的、可直接用于机器学习的数据集。该数据集源自瑞士住宅数据库（Swiss Dwellings，版本3.0.0）。

MSD数据集的训练集划分包含4167张覆盖瑞士全境的单户及多户建筑综合体户型图，相比RPLAN数据集（RPLAN dataset）等知名户型图数据集，其涵盖的建筑规模更广。由于MSD数据集将作为2023年10月3日于巴黎举办的国际计算机视觉大会（International Conference on Computer Vision，ICCV）赛事的一部分，其测试集划分尚未公开。该测试集将在赛事投稿截止日期（约2023年9月中旬）过后发布。

## 数据清洗、筛选与预处理
所有清洗、筛选及预处理工作均通过Python语言完成。研究人员首先对瑞士住宅数据库进行清洗与筛选，仅保留房间数量不少于10间、且至少包含2个“2区房间”（例如客厅、走廊、厨房、餐厅）的住宅综合体数据。研发了一套完全基于Python的`shapely`与`networkx`库的图提取算法，用于从筛选后的户型图中提取通行拓扑图。

## 数据集结构
MSD数据集包含3个文件：
1. README.md文件：用于说明数据集相关信息。
2. 训练集压缩包：内含4个文件夹，分别为`graph_in`（存储格式为<index>.pickle）、`struct_in`（存储格式为<index>.npy）、`full_out`（存储格式为<index>.npy）与`graph_out`（存储格式为<index>.pickle）。所有文件夹内的文件命名规则保持一致，即命名为<index>.pickle的`graph_in`文件与命名为<index>.npy的`full_out`文件属于同一户型图数据。
3. 测试集压缩包：内含`graph_in`与`struct_in`两个文件夹，结构与训练集数据一致，但未公开标注信息。

## 户型图自动补全任务
MSD数据集的研发目标是助力计算机科学领域研究者开发面向户型图自动补全任务的深度学习模型。户型图自动补全任务的输入内容包括建筑边界、保障建筑结构安全所需的结构元素，以及以图结构形式定义的一组用户约束条件，目标是自动生成完整的户型图。具体而言，该任务的目标是学习`graph_in`与`struct_in`的联合分布与`full_out`分布之间的关联关系。若研究者需要使用或开发图信号处理、专用图机器学习相关方法，则可使用`graph_out`文件。

## GitHub使用指南
相关GitHub页面。