Multifaceted Exploration of Spatial Openness in Rental Housing: A Big Data Analysis in Tokyo's 23 Wards

Understanding spatial openness is vital for improving residential quality and design; however, studies often treat its influencing factors separately. This study developed a quantitative framework to evaluate the spatial openness in housing from two- (2D) and three- (3D) dimensional perspectives. Using data from 4,004 rental units in Tokyo’s 23 wards, we examined the temporal and spatial variations in openness and its relationship with rent and housing attributes. 2D openness was computed via planar visibility using visibility graph analysis (VGA) from floor plans, whereas 3D openness was derived from interior images analysed using Mask2Former, a semantic segmentation model that identifies walls, ceilings, floors, and windows. The results showed an increase in living room visibility and a 1990s peak in overall openness. Spatial analyses revealed partial correlations among openness, rent, and building characteristics, reflecting urban redevelopment trends. Although the 2D and 3D openness indicators were not directly correlated, higher openness tended to correspond to higher rent. The impression scores predicted by the existing models were only weakly related to openness, suggesting that the interior design and furniture more strongly shape perceived space. This study offers a new multidimensional data-driven framework for quantifying residential spatial openness and linking it with urban and market dynamics.

💡 Research Summary

**
This paper presents a novel, data‑driven framework for quantifying spatial openness in residential units from both two‑dimensional (2D) planar visibility and three‑dimensional (3D) interior composition perspectives. Using a large‑scale dataset of 4,004 rental apartments located in Tokyo’s 23 wards, the authors compute 2D openness through Visibility Graph Analysis (VGA) applied to floor‑plan images, and 3D openness through semantic segmentation of interior photographs using the state‑of‑the‑art Mask2Former model.

Data collection and preprocessing
The dataset comprises properties built after 1960, each with a high‑resolution floor plan, at least one interior photo of the living room, and metadata such as construction year, floor area, location, and monthly rent. Properties lacking a usable interior image were excluded, resulting in a clean sample of 4,004 units.

2D openness computation
Floor‑plan images are first processed by a DeepLab‑V3+ network to label walls, doors, and windows at the pixel level. After binary conversion (walls = 1, everything else = 0) and scaling to the actual floor area, a uniform 20 cm grid is overlaid. For each grid point the algorithm counts how many other points can be connected by a straight line without intersecting a wall, thereby constructing a visibility graph. The average number of visible connections per grid point constitutes the 2D openness score. This automated pipeline extends traditional space‑syntax methods (isovists, VGA) to thousands of units without manual graph construction.

3D openness computation
Interior photographs are fed into Mask2Former, pretrained on the ADE20K dataset, to segment four semantic classes: ceiling, floor, wall, and window. The pixel count of each class is divided by the total number of visible pixels (i.e., the sum of all four classes) to obtain proportionate 3D openness metrics. Although photographs were taken from heterogeneous viewpoints, lighting conditions, and depths, the large sample size mitigates systematic bias, allowing the authors to capture average compositional trends across the city.

Temporal trends
Analyzing openness over construction year reveals a pronounced peak in 2D visibility scores in the early 1990s, followed by a gradual decline. In the 3D domain, the proportion of ceiling area has risen steadily since the 1990s, whereas window proportion has decreased. The authors interpret these patterns as the result of large‑scale redevelopment in the 1990s that emphasized open‑plan living rooms and higher ceilings, while later design trends have favored more compact window arrangements, possibly due to energy‑efficiency concerns.

Spatial distribution
Geographically, central wards exhibit higher 2D openness and larger window ratios, reflecting dense, high‑rise environments where visual connectivity is a design priority. Peripheral wards show lower 2D openness, driven by housing types with more subdivided rooms and larger floor areas. This spatial heterogeneity aligns with known urban redevelopment trajectories and differing housing typologies across Tokyo.

Correlation with rent and other variables
Statistical analysis demonstrates that 2D and 3D openness metrics are largely independent (no significant correlation). Nevertheless, both metrics display positive associations with monthly rent. Specifically, a one‑standard‑deviation increase in the 2D openness score corresponds to an approximate 8 % rise in average rent, after controlling for floor area, building age, and location. This finding corroborates prior research linking visual openness to perceived residential value. Conversely, the “impression scores” derived from existing housing‑preference models show only weak correlations with either openness metric, suggesting that occupants’ subjective experience is more strongly shaped by interior furnishings, lighting, and color schemes than by pure spatial geometry.

Limitations
The authors acknowledge several constraints: (1) interior photographs vary in angle, distance, and illumination, potentially biasing 3D measurements; (2) floor plans may not perfectly reflect as‑built interiors, especially after renovations; (3) the sample is confined to Tokyo’s 23 wards, limiting external generalizability.

Implications and future work
The study contributes a robust, scalable methodology for assessing spatial openness that can be applied wherever floor plans and interior images are available. Potential applications include: (a) informing architects and planners on how design choices affect market value; (b) enriching real‑estate recommendation engines with openness‑based personalization; (c) supporting urban policy by linking openness trends to redevelopment cycles. Future research should aim to (i) collect multi‑view, high‑resolution interior imagery to reduce segmentation bias; (ii) integrate sensor‑based spatial data (e.g., LiDAR, depth cameras) for more precise 3D modeling; and (iii) extend the analysis to other metropolitan areas to test the framework’s universality.

In sum, by jointly quantifying 2D visibility and 3D compositional openness across a city‑wide rental market, the paper demonstrates that spatial openness is a measurable, market‑relevant attribute that varies over time and space, and that its influence on rent operates independently of traditional impression‑based preference scores. This multidimensional, big‑data approach opens new avenues for evidence‑based residential design and urban planning.