Dataset Collection by Group of Photogrammetry and Computer Vision (GPCV) at Wuhan University

Shunping Ji

Dataset

Dataset 1: WHU Building Dataset

Summary: The dataset consists of an aerial image sub-dataset, two satellite image sub-datasets and a building change detection sub-dataset covering more than 1400 km2.

Paper: Fully Convolutional Networks for Multi-Source Building Extraction from An Open Aerial and Satellite Imagery Dataset.


Dataset 2: Omnidirectional Street-View (OSV) Dataset for Spherical Object Detection

Summary: We create an omnidirectional image dataset of real street scenes called OSV dataset with multi-class annotations for spherical object detection. It was collected by a vehicle-mounted panoramic camera and contains 1777 lights, 867 cars, 578 traffic signs, 867 crosswalks and 355 crosswalk warning lines, totally 5636 objects.

Paper: Grid based Spherical CNN for Object Detection from Panoramic Images.


Dataset 3: GF2 Dataset for 3DFGC

Summary: Multi-temporal multi-spectral GF2 satellite image dataset for crop type classification along with the 3D FCN with Global pooling module and Channel attention module (3DFGC) source code.

Paper: Learning discriminative spatiotemporal features for precise crop classification from multi-temporal satellite images.


Dataset 4: Bijie Landslide Dataset

Summary: We create an open remote sensing landslide dataset called Bijie landslide dataset for developing automatic landslide detection methods. The dataset consists of satellite optical images, shapefiles of landslides’ boundaries and digital elevation models. All the images in this dataset, i.e. 770 landslide images (red points) and 2003 non-landslide images were cropped from the TripleSat satellite images captured from May to August 2018.

Paper: Landslide detection from an open satellite imagery and digital elevation model dataset using attention boosted convolutional neural networks.


Dataset 5: WHU Cloud Dataset

Summary: We manually edited a Landsat 8 dataset for cloud detection and removal, which contains the cloudy images, corresponding cloudless historical images, and cloud and shadow masks.

Paper: Simultaneous Cloud Detection and Removal From Bitemporal Remote Sensing Images Using Cascade Convolutional Neural Networks.


Dataset 6: WHU MVS/Stereo dataset

Summary: We created the synthetic aerial dataset for large-scale Earth surface reconstruction called the WHU dataset.

Paper: A Novel Recurrent Encoder-Decoder Structure for Large-Scale Multi-view Stereo Reconstruction from An Open Aerial Dataset.


Dataset 7: Ningbo Pylon dataset

Summary: We edited and released a large-scale electricity pylon detection dataset, the first open-source aerial pylon dataset called the Ningbo Pylon dataset.

Paper: A New Spatial-Oriented Object Detection Framework for Remote Sensing Images.


Dataset 8: WHU TLC dataset

Summary: The dataset is public for MVS (Multi-View Stereo) task in satellite domain, which is consisted of the triple-view satellite images, the RPC parameters and the ground-truth DSMs.

Paper: Rational Polynomial Camera Model Warping for Deep Learning Based Satellite Multi-View Stereo Matching.


Dataset 9: Maduo Earthquake Crack Dataset

Summary: We edited and released a high-resolution earthquake crack dataset, the first open-source aerial earthquake crack dataset called the Maduo dataset.

Paper: Earthquake Crack Detection from Aerial Images Using a Deformable Convolutional Neural Network.


Dataset 10: WHU-Mix (raster) building dataset

Summary: The WHU-Mix (raster) dataset is a diverse, large-scale, and high-quality dataset that aims to better simulate the situation of practical building extraction, to measure more reasonably the real performance of a deep learning model, and to evaluate more conveniently the generalization ability of a model on different remote sensing images acquired from different sources and places.

Paper: A diverse large-scale building dataset and a novel plug-and-play domain generalization method for building extraction.


Dataset 11: WHU-Mix (vector) building dataset

Summary: The WHU-Mix dataset consists of 64k image tiles with over 754k buildings, and covers an area of about 1100 km2. Nevertheless, the WHU-Mix dataset is mainly intended for segmentation-based methods; however, in this work, the focus is on direct vector format building polygon extraction. To adapt to this, we pulled the corresponding raw data and after editing, named it the WHU-Mix (vector) dataset. The WHU-Mix (vector) dataset uses the MS-COCO format for the building labels.

Paper: BuildMapper: A Fully Learnable Framework for Vectorized Building Contour Extraction.


Dataset 12: Multi-Temporal Sentinel-1/2 (MTS12) Dataset for land cover classification

Summary: The MTS12 dataset named as is composed of optical Sentinel-2 and SAR Sentinel-1 time series from January to December 2019 covering the whole Slovenia, an area of 20,271 square kilometers, for land cover classification. The whole dataset included 936 patches, each with the size of 500×500 pixels and the temporal dimension of 12 (one per month), consisting of eight land-cover classes, which are cultivated land, forest, grassland, shrubland, water, wetland, artificial surface, and bare land.

Paper: Linying Zhao & Shunping Ji, CNN, RNN, or ViT? An evaluation of different deep learning architectures for spatio-temporal representation of Sentinel time series, JSTARS, 2022.


Dataset 13: WHU-OMVS Dataset

Summary: We built a synthetic oblique aerial image MVS dataset (the WHU-OMVS dataset) for the purpose of city-level 3D scene reconstruction. The WHU-OMVS dataset was rendered from the same base scene as the WHU-MVS dataset (Liu and Ji, 2020), extending the original nadir-view dataset to an oblique five-view stereo dataset.

Paper: Liu Jin, Gao jian, Ji Shunping, Zeng Chang, Zhang Shaoyi, Gong jianya. Deep learning based multi-view stereo matching and 3D scene reconstruction from oblique aerial images. ISPRS Journal of Photogrammetry and Remote Sensing, 2023.



Code

Code (1):Attention boosted bilinear pooling for remote sensing image retrieval

Summary: We introduce a second-order pooling named compact bilinear pooling (CBP) into convolutional neural networks (CNNs) for remote sensing image retrieval.

Paper: Attention boosted bilinear pooling for remote sensing image retrieval.


Code (2) : Panoramic SLAM from a multiple fisheye camera rig

Summary: We develop a feature-based simultaneous localization and mapping (SLAM) system for panoramic image sequences obtained from a multiple fisheye camera rig in a wide baseline mobile mapping system (MMS).

Paper: Panoramic SLAM from a Multiple Fisheye Camera Rig.


Code (3): Matching RGB and Infrared Remote Sensing Images with Densely-Connected Convolutional Neural Networks

Summary: We develop a deep learning-based matching method between an RGB and an infrared image that were captured from satellite sensors.

Paper: Matching RGB and Infrared Remote Sensing Images with Densely-Connected Convolutional Neural Networks.


Code (4): Generative adversarial network-based full-space domain adaptation for land cover classification from multiple source remote sensing images

Summary: We propose a novel end-to-end GAN-based full-space domain adaptation learning framework.

Paper: Generative adversarial network-based full-space domain adaptation for land cover classification from multiple source remote sensing images.


Code (5): Automatic 3D building reconstruction from multi-view aerial images with deep learning

Summary: We develop a fully automatic method to reconstruct the LoD1 building from multi-view aerial images.

Paper: Automatic 3D building reconstruction from multi-view aerial images with deep learning.


Code (6): Multi-Scale Attentive Aggregation for LiDAR Point Cloud Segmentation

Summary: We proposed a multi-scale attentive aggregation network for semantic segmentation of LiDAR point cloud.

Paper: Multi-Scale Attentive Aggregation for LiDAR Point Cloud Segmentation.


Code (7): Simultaneous Road Surface and Centerline Extraction from Large-Scale Remote Sensing Images Using CNN-Based Segmentation and Tracing

Summary: A novel CNN-based multistage framework is proposed for simultaneous road surface and centerline tracing from remote sensing images instead of treating them separately as most current road extraction methods do.

Paper: Simultaneous Road Surface and Centerline Extraction from Large-Scale Remote Sensing Images Using CNN-Based Segmentation and Tracing.


Code (8): Scribble-Based Weakly Supervised Deep Learning for Road Surface Extraction from Remote Sensing Images

Summary: We propose a scribble-based weakly supervised road surface extraction method named ScRoadExtractor, which learns from easily accessible scribbles such as centerlines instead of densely annotated road surface ground-truths.

Paper: Paper: Scribble-Based Weakly Supervised Deep Learning for Road Surface Extraction from Remote Sensing Images.


Code (9): An End-to-End Contour-based Method for High-Quality High-Speed Instance Segmentation

Summary: We introduce a end-to-end contour-based instance segmentation method named E2EC, which achieves SOTA performance on SBD, KINS, COCO and Cityscapes dataset.

Paper: Paper: E2EC: An End-to-End Contour-based Method for High-Quality High-Speed Instance Segmentation.


Code (10): A concentric loop convolutional neural network for manual delineation level building boundary segmentation from remote sensing images (Under maintenance)

Summary: We propose a concentric loop convolutional neural network (CLP-CNN) method for the automatic segmentation of building boundaries from remote sensing images.

Paper: A concentric loop convolutional neural network for manual delineation level building boundary segmentation from remote sensing images.


Code (11): A general deep learning based framework for 3D reconstruction from multi-view stereo satellite images

Summary: We propose a general deep learning based framework, named Sat-MVSF, to perform three-dimensional (3D) reconstruction of the Earth’s surface from multi-view optical satellite images.

Paper: A general deep learning based framework for 3D reconstruction from multi-view stereo satellite images.


Code (12): Long-Range Correlation Supervision for Land-Cover Classification from Remote Sensing Images

Summary: we propose a novel supervised long-range correlation method for land-cover classification, called the supervised longrange correlation network (SLCNet), which is shown to be superior to the currently used unsupervised strategies.

Paper: Long-Range Correlation Supervision for Land-Cover Classification from Remote Sensing Images.

Code address: https://github.com/yudawen/SLCNet