From street view imagery to the countryside: large-scale perception of rural China using deep learning

This project includes the deep learning model code about the article "From street view imagery to the countryside: large-scale perception of rural China using deep learning," which introduces a Pair-CNN model. This model is applied to rural street view data to evaluate the subjective perceptions of rural China on five dimensions: Wealthy, Lively, Habitable, Tidy, and Terroir. The project includes both the data and code that support the Pair-CNN model.1. "data" folderPicture.zip—Rural street view imagery data, containing 100 randomly selected rural imagery in JPG format.label_fy.csv—Label data, recording volunteers' comparisons of two rural images based on the wealth metric (only includes images from Picture.zip). It has three fields: lefturl, righturl, and label. The label field is a binary classification result: 0 means the left image is better than the right image, and 1 means the opposite.2. "model" folderResNet.py—The basic structure of the model, using ResNet50ComparaNet.py—The basic structure of the Pair-CNN model3."other" folderGet_Image.py—Script for loading image dataGridMask.py—Data augmentation4."run" folderTrain_Compare.ipynb—Executable file for model trainingTrueSkill.ipynb—Executable file for the TrueSkill algorithm5.requirements.txtEnvironment configuration and version numbers, including Python 3.9 and Pytorch 2.2