RSIBE: A large-scale dataset for extracting buildings from remote sensing images (Jichong Yin, Fang Wu, Haizhong Qian, Yue Qiu, Chengyi Liu, Xianyong Gong and Andong Wang) Continuous updating

This repository is the official implementation of RSIBE: A large-scale dataset for extracting buildings from remote sensing images by Jichong Yin, Fang Wu, Haizhong Qian, Yue Qiu, Chengyi Liu, Xianyong Gong and Andong Wang.

Instances of the images in the RSIBE dataset captured in different imaging conditions, weathers, seasons, and image qualities, including various types and sizes of the buildings in different areas.

Multi-format and high-quality sample annotation: (a) Image; (b) Annotation of the VOC format; (c) Annotation of the vector format; (d) Annotation of the COCO format for building contour segmentation; (e) Annotation of the COCO format for building object detection; (f) Annotation of the Labelme format; (g) The binary label; (h) Annotation of the building boundary.

Samples of the RSIBE dataset with various scales.

Samples of the RSIBE dataset with significant intra-class diversity.

Comparison of the results at the overall level.

Comparison of the results at the regional level.

Comparison of results at the building level.

Comparison of transfer learning results using the RSIBE dataset.

Building instance segmentation using YOLACT on the RSIBE dataset: (a) Image; (b) Label;

Building boundary extraction using HRNet on the RSIBE dataset.