Datasets:

junhao-zhang
/

InSpect

+{
+  "@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/",
+    "prov": "http://www.w3.org/ns/prov#",
+    "equivalentProperty": "cr:equivalentProperty",
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "samplingRate": "cr:samplingRate"
+  },
+  "@type": "sc:Dataset",
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "metadata-csv",
+      "name": "specimen_benchmark_metadata.csv",
+      "description": "Main specimen metadata table: one row per specimen image, including image name, hierarchical taxonomy, label-derived metadata, taxon keys, and benchmark split assignments.",
+      "contentUrl": "https://huggingface.co/datasets/junhao-zhang/InSpect/resolve/main/specimen_benchmark_metadata.csv",
+      "encodingFormat": "text/csv",
+      "contentSize": "32069172 B",
+      "sha256": "e39ca855ab86fbedade2848676e4a96e0984dc8f6592a6e45bcf193299a2fcfb"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "metadata-jsonl",
+      "name": "specimen_benchmark_metadata.jsonl",
+      "description": "JSONL representation of specimen benchmark metadata.",
+      "contentUrl": "https://huggingface.co/datasets/junhao-zhang/InSpect/resolve/main/specimen_benchmark_metadata.jsonl",
+      "encodingFormat": "application/x-ndjson",
+      "contentSize": "57801937 B",
+      "sha256": "2a00ff06a1f4cffe636840dac444e9bec10e135a2ff95446bb6e1e1436698308"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "original-images",
+      "name": "original_images.zip",
+      "description": "Original collection photographs containing insect specimens and their physical specimen labels.",
+      "contentUrl": "https://huggingface.co/datasets/junhao-zhang/InSpect/resolve/main/original_images.zip",
+      "encodingFormat": "application/zip",
+      "contentSize": "248146898186 B",
+      "sha256": "2cb091b495298ce9e013f5cffb58474fbe51f20f43b9937eca218e0aebb543f3"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "specimen-crops",
+      "name": "final_benchmark_data_crops.zip",
+      "description": "Aligned cropped insect images used for specimen recognition.",
+      "contentUrl": "https://huggingface.co/datasets/junhao-zhang/InSpect/resolve/main/final_benchmark_data_crops.zip",
+      "encodingFormat": "application/zip",
+      "contentSize": "17500289692 B",
+      "sha256": "e661135b3668a55dd1476908bb4b0a6e07c16d68a9b1c5fc67fb868bb673d760"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "segmentation-images",
+      "name": "segmentation_images.zip",
+      "description": "Cropped images for the curated anatomical part segmentation subset.",
+      "contentUrl": "https://huggingface.co/datasets/junhao-zhang/InSpect/resolve/main/segmentation_images.zip",
+      "encodingFormat": "application/zip",
+      "contentSize": "801912896 B",
+      "sha256": "7e4b87c469d8d75647c6da27444a1582e8087b23a88c2e5fca4d1a7af792916b"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "segmentation-annotations",
+      "name": "segmentation_annotations.zip",
+      "description": "COCO-format anatomical part annotations and train, validation, and test splits for singular and spatial-aware labels. The two label settings use different split assignments.",
+      "contentUrl": "https://huggingface.co/datasets/junhao-zhang/InSpect/resolve/main/segmentation_annotations.zip",
+      "encodingFormat": "application/zip",
+      "contentSize": "13956172 B",
+      "sha256": "6cdca64e334a45102697c446f8557637cf5e2572de01c2e9c8250742922d5853"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "metadata",
+      "description": "One record per specimen image in specimen_benchmark_metadata.csv; 48,184 records. Field names and types follow the released metadata table.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/image_name",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "image_name"
+            }
+          },
+          "name": "image_name"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/catalog_number",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "catalog_number"
+            }
+          },
+          "name": "catalog_number"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/scientific_name",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "scientific_name"
+            }
+          },
+          "name": "scientific_name"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/order",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "order"
+            }
+          },
+          "name": "order"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/family",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "family"
+            }
+          },
+          "name": "family"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/genus",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "genus"
+            }
+          },
+          "name": "genus"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/species",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "species"
+            }
+          },
+          "name": "species"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/authority",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "authority"
+            }
+          },
+          "name": "authority"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/year",
+          "dataType": "cr:Float64",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "year"
+            }
+          },
+          "name": "year"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/locality",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "locality"
+            }
+          },
+          "name": "locality"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/original_locality",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "original_locality"
+            }
+          },
+          "name": "original_locality"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/state_province",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "state_province"
+            }
+          },
+          "name": "state_province"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/verbatim_coordinates",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "verbatim_coordinates"
+            }
+          },
+          "name": "verbatim_coordinates"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/latitude",
+          "dataType": "cr:Float64",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "latitude"
+            }
+          },
+          "name": "latitude"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/longitude",
+          "dataType": "cr:Float64",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "longitude"
+            }
+          },
+          "name": "longitude"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/collection_date",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "collection_date"
+            }
+          },
+          "name": "collection_date"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/collector",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "collector"
+            }
+          },
+          "name": "collector"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/sampling_event",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "sampling_event"
+            }
+          },
+          "name": "sampling_event"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/life_stage",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "life_stage"
+            }
+          },
+          "name": "life_stage"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/sex",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "sex"
+            }
+          },
+          "name": "sex"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/determiner",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "determiner"
+            }
+          },
+          "name": "determiner"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/determination_date",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "determination_date"
+            }
+          },
+          "name": "determination_date"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/institution",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "institution"
+            }
+          },
+          "name": "institution"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/notes",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "notes"
+            }
+          },
+          "name": "notes"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/notes_clean",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "notes_clean"
+            }
+          },
+          "name": "notes_clean"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/weak_context_text",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "weak_context_text"
+            }
+          },
+          "name": "weak_context_text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/label_text",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "label_text"
+            }
+          },
+          "name": "label_text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/taxon_key",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "taxon_key"
+            }
+          },
+          "name": "taxon_key"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/split",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "split"
+            }
+          },
+          "name": "split",
+          "description": "Benchmark split assignment from the released metadata CSV (train, val, test_seen, or test_unseen); this is not the Hugging Face viewer configuration split."
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "metadata/species_split",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "metadata-csv"
+            },
+            "extract": {
+              "column": "species_split"
+            }
+          },
+          "name": "species_split"
+        }
+      ],
+      "name": "specimen_benchmark_metadata"
+    }
+  ],
+  "conformsTo": "http://mlcommons.org/croissant/1.1",
+  "version": "1.0.0",
+  "name": "InSpect",
+  "description": "InSpect: A Curated Natural History Collection Dataset for Insect Specimen Understanding. The dataset contains 48,184 digitized insect specimen images from the Australian National Insect Collection (ANIC), spanning 653 taxa, with aligned insect crops, hierarchical taxonomic labels, and label-derived structured metadata. A curated subset of 1,859 images includes fine-grained anatomical part masks in singular and spatial-aware label settings. The benchmarks cover open taxonomic recognition and supervised and text-guided anatomical part segmentation. Metadata-assisted recognition uses specimen metadata during fine-tuning; inference uses specimen images without per-specimen metadata.",
+  "alternateName": [
+    "junhao-zhang/InSpect"
+  ],
+  "creator": [
+    {
+      "@type": "sc:Person",
+      "name": "Junhao Zhang"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Feng Xu"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Ahalya Ravendran"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Nicole Fisher"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Brendan Tidd"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Lars Petersson"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Dadong Wang"
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Xun Li"
+    }
+  ],
+  "keywords": [
+    "insects",
+    "biodiversity",
+    "natural-history-collections",
+    "taxonomy",
+    "image-classification",
+    "zero-shot-image-classification",
+    "image-segmentation",
+    "open-taxonomic-recognition",
+    "anatomical-part-segmentation",
+    "cc-by-nc-nd-4.0"
+  ],
+  "license": "https://creativecommons.org/licenses/by-nc-nd/4.0/",
+  "url": "https://huggingface.co/datasets/junhao-zhang/InSpect",
+  "rai:dataLimitations": "InSpect is built from preserved insect specimens from the Australian National Insect Collection, so it reflects the taxonomic, geographic, temporal, imaging, and curation scope of this collection. It is designed for digitized museum specimen images, not live insects, field photographs, or other collections without additional validation. The dataset is long-tailed, crop quality varies, metadata is incomplete and noisy, and fine-grained part masks are available only for a curated subset. It should not be used for automated taxonomic decisions, commercial identification, ecological deployment, or conservation decisions without expert review.",
+  "rai:dataBiases": "InSpect may contain taxonomic, geographic, temporal, and collection-specific biases from its source collection. Some taxa are much more represented than others, and images were selected for identifiable species names and sufficient visual quality. Metadata fields are unevenly available and may encode collection-specific correlations such as locality, collector, or time period. Models trained on this dataset may learn these dataset-specific patterns rather than robust biological cues.",
+  "rai:personalSensitiveInformation": "The dataset does not contain human-subject images, surveillance data, health data, political or religious information, or protected human attributes. Some metadata fields may include names of taxonomic authorities, collectors, or determiners from specimen labels; these are treated as scientific provenance metadata and should not be used to profile individuals. The dataset may include locality or coordinate information, which can be sensitive for rare or vulnerable species.",
+  "rai:dataUseCases": "InSpect is intended for research on digitized insect specimen understanding, open taxonomic recognition, label-derived metadata analysis, and fine-grained anatomical part segmentation. It has been validated through benchmark experiments on zero-shot/fine-tuned recognition and supervised/text-guided segmentation. It has not been validated for live insects, field images, commercial identification, automated taxonomy, or conservation decision-making without expert review.",
+  "rai:dataSocialImpact": "InSpect can support biodiversity informatics, natural history collection digitization, taxonomic curation, specimen organization, preliminary taxonomic screening, and morphology-level analysis. Potential negative impacts include misleading automated identifications, overgeneralization beyond the source collection, and inappropriate use of incomplete or noisy metadata. Collection biases may lead to uneven performance across taxa and regions. Collector and determiner names require appropriate care, and locality information may be sensitive for rare or vulnerable species. Users should follow applicable institutional policies and relevant regulations, use expert review for taxonomic or conservation decisions, and consider efficient training and evaluation to reduce computational and environmental costs. Further discussion is provided in Appendix A of the accompanying paper.",
+  "rai:hasSyntheticData": false,
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://www.csiro.au/en/about/facilities-collections/Collections/ANIC",
+      "prov:label": "Australian National Insect Collection (ANIC)",
+      "prov:wasAttributedTo": {
+        "@id": "https://www.csiro.au/",
+        "prov:label": "CSIRO / Australian National Insect Collection"
+      }
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "ANIC specimen image collection",
+      "sc:description": "Digitized preserved insect specimen images were obtained from the Australian National Insect Collection (ANIC). The source images were captured under indoor collection imaging conditions and contain single insect specimens together with their physical specimen labels.",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:Agent",
+          "@id": "australian_national_insect_collection_/_csiro",
+          "prov:label": "Australian National Insect Collection / CSIRO",
+          "sc:description": "Source collection provider"
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q1172378"
+      },
+      "prov:label": "Image filtering and insect cropping",
+      "sc:description": "We filtered the source images to retain specimens with identifiable species-level names and sufficient visual quality. Insect regions were detected using Grounding DINO and manually inspected to obtain accurate aligned crops. Recognition and segmentation benchmarks use the cropped insect images to separate visual morphology from physical label text."
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q109719325"
+      },
+      "prov:label": "Anatomical part mask annotation",
+      "sc:description": "A curated subset of 1,859 images was annotated with fine-grained anatomical part masks using X-AnyLabeling with SAM-assisted manual refinement. Two annotators annotated and reviewed masks for structures such as head, thorax, abdomen, antennae, legs, wings, and other visible parts."
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q109719325"
+      },
+      "prov:label": "Label-derived metadata extraction",
+      "sc:description": "Specimen-label information was extracted from the original collection images using gemma-4-31B-it with a structured prompt. The output was normalized into a fixed key-value schema containing catalog number, taxonomy, locality, coordinates, collection event, biological attributes, determination history, institution, and notes. Missing fields were represented as null."
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q3306762"
+      },
+      "prov:label": "Manual review and correction of annotations and extracted metadata",
+      "sc:description": "Metadata review and correction involved comparing extracted specimen-label fields with the original collection images and correcting identified errors. SAM-assisted anatomical part annotations were manually inspected and refined to improve part labels and mask boundaries."
+    }
+  ],
+  "citeAs": "Junhao Zhang, Feng Xu, Ahalya Ravendran, Nicole Fisher, Brendan Tidd, Lars Petersson, Dadong Wang, and Xun Li. InSpect: A Curated Natural History Collection Dataset for Insect Specimen Understanding. 2026."
+}