Green roof segmentation dataset

Our dataset consists of 300 training images and 180 test images, each cropped to 300 \u00d7 300 pixels. To assess the generalization ability of GR-Net in large-scale applications, the test dataset is divided into in-domain and out-of-domain subsets. The in-domain test dataset contains 100 images from Nanshan district (65 images) and Futian district (35 images), which are from the same regions as the training dataset. The out-of-domain test dataset includes 80 images from Luohu district.