{ "@context": { "@language": "en", "@vocab": "https://schema.org/", "arrayShape": "cr:arrayShape", "citeAs": "cr:citeAs", "column": "cr:column", "conformsTo": "dct:conformsTo", "containedIn": "cr:containedIn", "cr": "http://mlcommons.org/croissant/", "data": { "@id": "cr:data", "@type": "@json" }, "dataBiases": "cr:dataBiases", "dataCollection": "cr:dataCollection", "dataType": { "@id": "cr:dataType", "@type": "@vocab" }, "dct": "http://purl.org/dc/terms/", "extract": "cr:extract", "field": "cr:field", "fileProperty": "cr:fileProperty", "fileObject": "cr:fileObject", "fileSet": "cr:fileSet", "format": "cr:format", "includes": "cr:includes", "isArray": "cr:isArray", "isLiveDataset": "cr:isLiveDataset", "jsonPath": "cr:jsonPath", "key": "cr:key", "md5": "cr:md5", "parentField": "cr:parentField", "path": "cr:path", "personalSensitiveInformation": "cr:personalSensitiveInformation", "recordSet": "cr:recordSet", "references": "cr:references", "regex": "cr:regex", "repeated": "cr:repeated", "replace": "cr:replace", "sc": "https://schema.org/", "separator": "cr:separator", "source": "cr:source", "subField": "cr:subField", "transform": "cr:transform", "rai": "http://mlcommons.org/croissant/RAI/1.0" }, "@type": "sc:Dataset", "distribution": [ { "@type": "cr:FileObject", "@id": "repo", "name": "repo", "description": "The Hugging Face git repository.", "contentUrl": "https://huggingface.co/datasets/squid-bench-anon/SQuID/tree/refs%2Fconvert%2Fparquet", "encodingFormat": "git+https", "sha256": "https://github.com/mlcommons/croissant/issues/80" }, { "@type": "cr:FileSet", "@id": "parquet-files-for-config-default", "containedIn": { "@id": "repo" }, "encodingFormat": "application/x-parquet", "includes": "default/*/*.parquet" } ], "recordSet": [ { "@type": "cr:RecordSet", "dataType": "cr:Split", "key": { "@id": "default_splits/split_name" }, "@id": "default_splits", "name": "default_splits", "description": "Splits for the default config.", "field": [ { "@type": "cr:Field", "@id": "default_splits/split_name", "dataType": "sc:Text" } ], "data": [ { "default_splits/split_name": "train" } ] }, { "@type": "cr:RecordSet", "@id": "default", "description": "squid-bench-anon/SQuID - 'default' subset", "field": [ { "@type": "cr:Field", "@id": "default/split", "dataType": "sc:Text", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "fileProperty": "fullpath" }, "transform": { "regex": "default/(?:partial-)?(train)/.+parquet$" } }, "references": { "field": { "@id": "default_splits/split_name" } } }, { "@type": "cr:Field", "@id": "default/id", "dataType": "sc:Text", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "id" } } }, { "@type": "cr:Field", "@id": "default/image", "dataType": "sc:ImageObject", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "image" }, "transform": { "jsonPath": "bytes" } } }, { "@type": "cr:Field", "@id": "default/question", "dataType": "sc:Text", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "question" } } }, { "@type": "cr:Field", "@id": "default/answer", "dataType": "sc:Text", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "answer" } } }, { "@type": "cr:Field", "@id": "default/type", "dataType": "sc:Text", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "type" } } }, { "@type": "cr:Field", "@id": "default/tier", "dataType": "cr:Int32", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "tier" } } }, { "@type": "cr:Field", "@id": "default/gsd", "dataType": "cr:Float32", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "gsd" } } }, { "@type": "cr:Field", "@id": "default/acceptable_range_lower", "dataType": "cr:Float64", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "acceptable_range_lower" } } }, { "@type": "cr:Field", "@id": "default/acceptable_range_upper", "dataType": "cr:Float64", "source": { "fileSet": { "@id": "parquet-files-for-config-default" }, "extract": { "column": "acceptable_range_upper" } } } ] } ], "conformsTo": "http://mlcommons.org/croissant/1.1", "name": "SQuID", "description": "\n\t\n\t\t\n\t\tSQuID: Satellite Quantitative Intelligence Dataset\n\t\n\nA comprehensive benchmark for evaluating quantitative spatial reasoning in Vision-Language Models using satellite imagery.\n\n\t\n\t\t\n\t\tDataset Overview\n\t\n\n\n2000 questions testing spatial reasoning on satellite imagery\n587 unique images across four datasets\n1950 auto-labeled questions from segmentation masks (DeepGlobe, EarthVQA, Solar Panels)\n50 human-annotated questions from NAIP imagery with consensus answers\n1577 questions include… See the full description on the dataset page: https://huggingface.co/datasets/squid-bench-anon/SQuID.", "alternateName": [ "squid-bench-anon/SQuID", "SQuID" ], "creator": { "@type": "Person", "name": "Anonymous Authors", "url": "https://huggingface.co/squid-bench-anon" }, "keywords": [ "visual-question-answering", "English", "cc-by-nc-4.0", "1K - 10K", "parquet", "optimized-parquet", "Image", "Text", "Datasets", "Dask", "Polars", "Croissant", "🇺🇸 Region: US", "satellite-imagery", "spatial-reasoning", "benchmark", "quantitative-reasoning", "VLM", "language-understanding" ], "license": "https://choosealicense.com/licenses/cc-by-nc-4.0/", "url": "https://huggingface.co/datasets/squid-bench-anon/SQuID", "dataCollection": "SQuID is constructed from four publicly available remote-sensing sources: EarthVQA (ICCV 2023, building/land-cover masks), DeepGlobe (CVPR Workshops 2018, land-use classification), a photovoltaic panel segmentation dataset (Earth System Science Data 2021), and NAIP imagery (USGS public domain). Auto-labeled questions (1,950) are derived programmatically from segmentation masks via geometric operations (area, count, distance, proximity) using OpenCV and scipy. Human-annotated questions (50) on NAIP imagery were collected from 10 independent annotators per question via a custom web interface, with median/majority voting for ground truth and Mean Median Absolute Deviation (MAD) for tolerance ranges.", "dataBiases": "Geographic bias: source datasets concentrate on US (NAIP, photovoltaic) and selected international urban/rural scenes (DeepGlobe, EarthVQA); SQuID is not representative of all global land-use distributions. Resolution bias: ground sampling distance restricted to 0.3m-1.0m; lower-resolution satellite imagery (e.g., Sentinel-2 at 10m) is out of scope. Class-vocabulary bias: questions are restricted to the closed set of land-use classes annotated in source datasets (e.g., building, forest, water, road, cropland, solar panel); open-vocabulary objects are not represented. Question-type bias: 24 question types reflect typical RS analytics tasks (counting, area, proximity, multi-condition); other geospatial reasoning forms (temporal change, 3D structure, semantic captioning) are out of scope.", "personalSensitiveInformation": "None. SQuID consists exclusively of overhead satellite/aerial imagery at 0.3m-1.0m ground sampling distance, which does not contain identifiable faces, license plates, or other personally identifiable information. No human subjects beyond voluntary annotators (whose identities are not stored) are involved.", "usageInfo": "SQuID is intended for academic research on quantitative spatial reasoning in vision-language models. Per its CC BY-NC 4.0 license (inherited from upstream LoveDA/EarthVQA terms), commercial use is prohibited.", "rai:dataCollection": "SQuID is constructed from four publicly available remote-sensing sources: EarthVQA (ICCV 2023, building/land-cover masks), DeepGlobe (CVPR Workshops 2018, land-use classification), a photovoltaic panel segmentation dataset (Earth System Science Data 2021), and NAIP imagery (USGS public domain). Auto-labeled questions (1,950) are derived programmatically from segmentation masks via geometric operations (area, count, distance, proximity) using OpenCV and scipy. Human-annotated questions (50) on NAIP imagery were collected from 10 independent annotators per question via a custom web interface, with median/majority voting for ground truth and Mean Median Absolute Deviation (MAD) for tolerance ranges.", "rai:dataBiases": "Geographic bias: source datasets concentrate on US (NAIP, photovoltaic) and selected international urban/rural scenes (DeepGlobe, EarthVQA); SQuID is not representative of all global land-use distributions. Resolution bias: ground sampling distance restricted to 0.3m-1.0m; lower-resolution satellite imagery (e.g., Sentinel-2 at 10m) is out of scope. Class-vocabulary bias: questions are restricted to the closed set of land-use classes annotated in source datasets (e.g., building, forest, water, road, cropland, solar panel); open-vocabulary objects are not represented. Question-type bias: 24 question types reflect typical RS analytics tasks (counting, area, proximity, multi-condition); other geospatial reasoning forms (temporal change, 3D structure, semantic captioning) are out of scope.", "rai:personalSensitiveInformation": "None. SQuID consists exclusively of overhead satellite/aerial imagery at 0.3m-1.0m ground sampling distance, which does not contain identifiable faces, license plates, or other personally identifiable information. No human subjects beyond voluntary annotators (whose identities are not stored) are involved.", "rai:dataUseCases": "SQuID is intended for academic research on quantitative spatial reasoning in vision-language models. Per its CC BY-NC 4.0 license (inherited from upstream LoveDA/EarthVQA terms), commercial use is prohibited.", "rai:dataCollectionType": "Machine-generated from curated segmentation masks with human annotation for validation and tolerance calibration", "rai:machineAnnotationTools": "OpenCV 8-connected component analysis, Euclidean distance transforms, and pixel-count area computation over curated segmentation masks; deterministic question templates with embedded minimum-area thresholds and GSD", "rai:annotationsPerItem": "10 human annotations per question on the 50-question NAIP validation subset (500 total); tolerance ranges derived via Median Absolute Deviation and applied benchmark-wide by question type", "rai:annotatorDemographics": "Volunteer research staff annotators; internal collection with no crowdsourcing and no compensation; annotator identities not stored", "rai:dataLimitations": "Questions restricted to mask-derivable quantities (no 3D/height, no multi-temporal change); GSD limited to 0.3m-1.0m; closed 9-class question vocabulary (urban, forest, agricultural, grass, vegetation, barren, water, solar, building); template-generated phrasing; proximity tolerances calibrated on fewer human items than counts/percentages (lower inter-rater reliability, Krippendorff alpha 0.42)", "rai:socialImpact": "Supports environmental monitoring, urban planning, and disaster response research by improving quantitative reliability of geospatial AI; overhead imagery at these resolutions contains no personally identifiable information", "rai:dataAnnotationProtocol": "10 independent annotations per question on the 50-question NAIP validation subset (500 annotations total, 11 annotators overall). Annotators completed a tutorial, then marked spatial regions, provided counts, or selected categorical answers. Per-question tolerance ranges were derived via Median Absolute Deviation (MAD) around the annotator consensus and applied benchmark-wide by question type.", "rai:dataAnnotationPlatform": "Custom grid-based annotation interface built on Turkle (Johns Hopkins University HLTCOE), with grid-cell region selection and a distance ruler; interface details in the supplementary material.", "rai:dataAnnotationAnalysis": "Inter-rater reliability: Krippendorff's alpha 0.79 overall (0.959 for counts, 0.424 for proximity). The released anonymized raw annotations and calibration script reproduce the published MAD, ICC and alpha values.", "rai:dataSocialImpact": "Supports environmental monitoring, urban planning, and disaster response research by improving quantitative reliability of geospatial AI; overhead imagery at these resolutions contains no personally identifiable information" }