The Role of Art Design in Public Space Renovation — A Study Based on Machine Learning Methods's date
Description
The database includes old photos of buildings in the "old building" folder, current photos of buildings in the "now building" folder, as well as various MATLAB programs for facade color analysis, along with specific analysis images and results (some results are described in Chinese). These are all stored in different folders.
Files
Steps to reproduce
Image preprocessing: Resize all facade images to 512×512 pixels and apply Gaussian filtering (σ=1.0). Deep feature extraction: Use the pre-trained ResNet-50 model (ImageNet weights) to extract 2048-dimensional features from the global average pooling layer. Color quantization: Apply K-means clustering (k=5) on the RGB color space of building area pixels to obtain dominant colors and their proportions. Style similarity: Calculate cosine similarity between the extracted feature vectors. Spatial scale: Use the ACF pedestrian detector to estimate building-to-pedestrian height ratios. Full methodological details are described in the accompanying research paper.