Geospatial dataset on human perceptions of wealth and physical disorder in urban China using street view imagery and deep learning.

Human perception is often considered a comprehensive evaluation of environmental quality. It is an important indicator of neighbourhood socioeconomic status and has a significant impact on various social outcomes. In response to the absence of locally trained models based on Chinese street view images and local annotators, we present two datasets. Dataset I, the perceived wealth and physical disorder scores annotation dataset, consists of 40,000 Chinese street view images that are annotated by local urban planners using the image comparison approach. Researchers can use Dataset I directly or further augment it to train their own artificial intelligence perception models in China. We use Dataset I to train image regression models, which are then employed to infer two perception scores for 36,262,700 street view images throughout urban China between 2013 and 2022. The resulting Dataset II, the perceived wealth and physical disorder scores prediction dataset, comprises three analytical units including image shooting points, 500m×500m grid cells, and 76,434 community administrative areas. Dataset II supports a variety of wide-coverage, fine-grained socio-spatial research projects in China, including studies on inequality, segregation, and gentrification. It can also serve as a critical input for examining other important socio-spatial phenomena, such as crime and physical activity patterns.