SoHo is a historic mixed-use neighborhood in Lower Manhattan, New York, best known for its cast-iron architecture, loft buildings, retail concentration, and intense pedestrian activity. Its streets combine landmarked nineteenth-century façades, large storefront windows, cobblestone blocks, and a dense mix of shopping, cultural, and residential uses, creating a visually rich street environment at eye level. This makes SoHo a productive case for streetscape analysis, because everyday urban experience here is strongly shaped by façade rhythm, storefront activity, pedestrian flow, and the changing balance between buildings, sidewalks, vehicles, greenery, and sky.
Panorama acquisition and data cleaning. The workflow started with spatial sampling along the street network, where observation points were prepared and linked to street-view imagery. Before analysis, the panorama dataset was cleaned to remove duplicate records and panoramas captured in indoor or enclosed environments, so that the final dataset would represent outdoor public streetscapes only. This step improved the consistency of the visual sample and reduced noise in the later semantic analysis.
Directional view processing. Instead of analysing the full panorama as a single continuous image, each panorama was divided into four directional views. This made the dataset more comparable across locations and allowed the study to capture front-facing, lateral, and reverse-facing street conditions separately. The split views provided a more controlled basis for semantic segmentation by reducing distortion and making street elements easier to read from a directional perspective.
Semantic segmentation and aggregation. Each of the four directional images was processed with a semantic segmentation model to classify visible urban elements such as buildings, sky, road, sidewalk, trees, vehicles, and people. After segmentation, the proportion of each class was calculated from pixel counts. The final indicator values used for the map markers were then aggregated from the four analysed directional images, producing one combined streetscape profile for each sampled location. This aggregation step made the markers represent the overall visual composition of a point rather than a single view only.
This map translates street-view imagery into measurable urban attributes. Instead of describing Soho only through parcels or land-use layers, the workflow captures eye-level environmental conditions and maps their spatial distribution.