WHU Roof10K dataset

The WHU-Roof10K dataset consists of 10,575 roof images derived from high-resolution remote sensing imagery covering diverse urban and suburban scenarios. Each image has a fixed size of 256*256 pixels and contains a single, complete building instance together with its immediate surroundings, ensuring sufficient contextual information while maintaining consistent input dimensions for network training.

For each image, the corresponding annotation provides the structural label of the rooftop, including the pixel coordinates of all corner points and their topological connectivity relationships.

The dataset is divided into a training set and a test set following a fixed partition specified by the train_list and test_list files, which provide the corresponding image filenames for each subset. Specifically, the training set contains 8,575 images, and the test set contains 2,000 images.