Polygon Similarity Benchmark Dataset

Dataset Description: Polygon Similarity Benchmark Dataset Overview This dataset provides a curated collection of polygonal shapes derived from real-world geographic information system (GIS) layers, specifically Parks, Water Bodies, and Sports categories from the SpatialHadoop GIS dataset.It is intended to support research in polygon representation learning, geometric similarity search, and spatial indexing. The dataset includes raw polygonal geometries, pre-computed similarity ground truth, and supplementary documentation. For each dataset category, 80% of the polygons were used to build the similarity index, while the remaining 20% were reserved exclusively for evaluation. Dataset Contents The distributed ZIP package contains the following files: 1. ShapeToVecResults2.pdf A supplementary document containing additional experimental results and visualizations referenced in the related publication. 2. PolygonSimilarityDataset.zip The main dataset archive containing all polygonal data, metadata, and ground-truth files. 3. poly_data.zip A collection of polygonal GIS datasets extracted from SpatialHadoop. These represent the input geometries used for similarity computation. 4. Ground Truth Files These archives contain precomputed shape similarity results for each domain: parks.tar sports.tar water_bodies.tar Each ground-truth archive consists of multiple text files, where each line represents a similarity query result. Ground Truth Format Each line in a ground-truth file encodes: <input_polygon_id> <similar_polygon_id_1> <similar_polygon_id_2> ... <similar_polygon_id_k> The first value is the ID of the input polygon. The subsequent values are the IDs of polygons determined to be most similar based on geometric shape similarity. The list of similar polygons is sorted in decreasing order of similarity, with the most similar polygon appearing first. These ground-truth lists were generated using geometric similarity metrics for evaluation and benchmarking of vector-based polygon encodings. Intended Use This dataset is primarily designed for: Research on polygon representation learning, embedding models, and shape encoders. Benchmarking approximate nearest-neighbor (ANN) algorithms on spatial shape data. Studying spatial indexing, vector search strategies, and geometric similarity measures. GIS analytics, spatial data mining, and machine learning applications involving polygonal geometries.