egrace479's picture
Update README stats following deduplication
e570b24
|
raw
history blame
18.3 kB
metadata
task_categories:
  - image-classification
tags:
  - butterfly
  - heliconius
  - dorsal
  - ventral
  - RGB
  - full body
  - separated wings
  - mimicry
  - CV
  - erato
  - melpomene
  - hybrids
  - cross types
  - wild
  - lab-bred
pretty_name: Jiggins Heliconius Collection
size_categories:
  - 10K<n<100K
language:
  - en
configs:
  - config_name: full_master
    data_files:
      - split: train
        path: Jiggins_Zenodo_Img_Master.csv
  - config_name: heliconius
    data_files:
      - split: train
        path: Jiggins_Heliconius_Master.csv
  - config_name: dorsal
    data_files:
      - split: train
        path: Jiggins_Zenodo_dorsal_Img_Master.csv

Dataset Card for Jiggins Heliconius Collection

Dataset Description

Dataset Summary

Subset of the collection records from Chris Jiggins' research group at the University of Cambridge, collection covers nearly 20 years of field studies. This subset contains approximately 36,211 RGB images of 11,968 specimens (29,134 images of 10,086 specimens across all Heliconius). Many records have both images and locality data.

Most images were photographed with a DSLR camera with a 100 mm macro-lens in standardized conditions. More information can be found at the individual Zenodo record pages. Images and full records with data are stored in the EarthCape database and on Zenodo (across 29 records from the Butterfly Genetics Group).

Both dorsal and ventral images available. Contains both whole butterfly, and 4 wings separate. Large variation in image content.

Languages

English, Latin

Dataset Structure

  • Jiggins_Zenodo_Img_Master.csv: Information for the approximately 36,0000 unprocessed image files included in the Jiggins Heliconius Collection. Image types are jpg, raw (.CR2) and tif. genus, species, and subspecies are included columns.

  • Jiggins_Zenodo_dorsal_Img_Master.csv: Subset of 17,759 images from Jiggins_Zenodo_Img_Master.csv with a dorsal view of the butterflies (note that some have both dorsal and ventral). This subset includes 11,752 unique specimens. Image types and columns are the same as for the Master file.

  • Jiggins_Heliconius_Master.csv: The 29,134-image subset of all Heliconius images from Jiggins_Zenodo_Img_Master.csv. This subset includes 10,086 unique specimens. Image types and columns are the same as for the Master file.

Notes:

  • The notebooks that generated these files and stats are included in the notebooks folder.
    • The original Jiggins Zenodo Master file was compiled from the CSVs provided with the included Zenodo records from the Butterfly Genetics Group. Christopher Lawrence selected which of these provided columns to include. Further processing and standardization (all documented in the Jupyter Notebooks) was performed by Elizabeth Campolongo.
    • Taxonomic information for records 5526257, 2554218, and 2555086 was recovered from information on their Zenodo pages, as the provided CSVs did not contain that information.
    • Be advised that there may be overlap between images in record 2548678 and record 3082688.
  • The scripts folder has a download and checksum script.
    • Images are downloaded to the provided images directory with subfolders labeled by the Taxonomic_Name, with filenames <X>_<Image_name>.
    • The checksum script is called by download_jiggins_subset.py to generate an MD5 for all downloaded images and creates a CSV with filepath, filename, and md5 columns in the same folder as the source CSV (named <source CSV>_checksums.csv). This helps to ensure FAIR and Reproducible results, though this will not distinguish between RAW and JPG versions of the same image.
    • A log of the download is also generated in the same folder as the source CSV (named <source CSV>_log.json).
  • metadata/Missing_taxa_Jiggins_Zenodo_Master.csv contains a record of the images that did not have easily reconcilable taxonomic information (see notebooks/standardize_datasets.ipynb for more information on this data). There are 1,630 such images distributed across 18 records.

Data Instances

Jiggins_Heliconius_Master.csv contains multiple species of Heliconius (including erato and melpomene), most are labeled down to the subspecies level. The Jiggins_Zenodo_Img_Master.csv also contains species from other genera, with just over half labeled to the subspecies level (these are predominantly Heliconius subspecies).

Detached wings in four quadrants (generally). Many include a label indicating the specimen ID (CAMID). There is variation in formatting both across and within records, but overall setup is relatively consistent. Some subspecies may be photographed differently, needs segmentation preprocessing.

  • Type: RAW (.CR2), JPG, and TIFF (very few)
  • Size (x pixels by y pixels): Unknown yet
  • Background (color or none): multiple (needs to be normalized, often grey or lime green)
  • Fit in frame: varies
  • Ruler or Scale: Some with Ruler
  • Color (ColorChecker, white-balance, None): some with ColorChecker, many with white reflectance standard in the bottom right corner.

Preprocessing steps (to be done):

  1. Hybrid separation - some images labeled as H. erato and H. melpomene without subspecies names are hybrids and need to be determined what subspecies they are.
  2. Label correction - along with step 1.
  3. Removal of subspecies with no mimic pairs.
  4. Segmentation of four wings from images so we can focus on forewings vs hindwings: WingSeg.

Current preprocessing steps: WingSeg on the Meier subset.

Data Fields

CSV Columns are as follows:

  • CAMID: Unique identifier for each specimen that was photographed. Each CAMID corresponds to multiple images (based on factors such as View and file_type).
  • X: Unique identifier for each line in the master CSV.
  • Image_name: Filename of image (unique, often CAM<CAMID>_<v or d>).
  • View: View of the butterfly in the image: dorsal, ventral, forewing dorsal, hindwing dorsal, forewing ventral, hindwing ventral, or dorsal and ventral.
  • zenodo_name: Name of the CSV file with metadata from the associated Zenodo record used to populate the information about this image.
  • zenodo_link: URL for the Zenodo record of the image.
  • Sequence: Mostly numeric IDs, not unique, please see the associated Zenodo record for more information on the meaning of these designations.
  • Taxonomic_Name: Indication of the Genus, species, and possibly, subspecies, of the specimen. For Cross Types, the hybrid names are reduced to just the two subspecies (from the Cross_Type column) and non-specified crosses are labeled <Genus> <species> cross hybrid.
  • Locality: Likely location of specimen collection, varying levels of specificity. Please see the associated Zenodo record for more information on the meaning of these designations.
  • Sample_accession: Some type of ID, not unique, please see the associated Zenodo record for more information on the meaning of these designations.
  • Collected_by: Abbreviations (likely for those collecting the specimen), please see the associated Zenodo record for more information on the meaning of these designations.
  • Other_ID: Mostly numeric IDs, not unique, please see the associated Zenodo record for more information on the meaning of these designations.
  • Date: Please see the associated Zenodo record for more information on the meaning of these designations.
  • Dataset: Overall collection the images belong to: Heliconiine Butterfly Collection Records from University of Cambridge (largely, but not entirely Heliconius), Patricio Salazar, Nadeau Sheffield, Bogota Collection (Camilo Salazar), Cambridge Collection, Mallet, Merril_Gamboa, STRI Collection (Owen). Please see the associated Zenodo record for more information on the meaning of these designations.
  • Store: Storage location for specimen (Box ###).
  • Brood: Likely indicator of batch of butterflies. Please see the associated Zenodo record for more information on the meaning of these designations.
  • Death_Date: Date of specimen death. Only noted for 269 images.
  • Cross_Type: Laboratory cross breeding information. There is a mix of F1 (subspecies x subspecies), F2 (F1 x F1), and backcross (F1 x subspecies) hybrids; these are all crosses of Heliconius erato and Heliconius melpomene subspecies. Generally, the order of the subspecies listed in the cross corresponds to the sexes of the parents (Maternal x Paternal). There are approximately 4,400 such images; on average, there are both raw and jpg images of the specimens for each view, so this covers 820 unique specimens.
  • Stage: Life stage of the specimen. Only 6 images (3 specimens) have a non-null value for this feature, and they are all labeled as Adult.
  • Sex: Sex of the specimen: Male, Female, or Unknown.
  • Unit_Type: Type of the specimen: wild, reared, Mutant, Wild, e11, or e12. Please see the associated Zenodo record for more information on the meaning of these designations.
  • file_type: Image type: jpg, raw (.CR2), or tif.
  • record_number: The number associated with the Zenodo record that the image came from.
  • species: Species of the specimen. There are 246 species represented in the full collection, 37 of these are species of Heliconius.
  • subspecies: Subspecies of the specimen (where available, mostly labeled for Heliconius). There are 155 subspecies represented in the full collection, 110 of which are Heliconius subspecies.
  • genus: Genus of the specimen. There are 94 unique genera represented in the full collection.
  • file_url: URL to download image from Zenodo: zenodo_link + "/files/" + Image_name. Allows for sample image display in data dashboard.
  • hybrid_stat: Hybrid status of the sample: hybrid, non-hybrid, or None. Hybrids are determined by an x or hybrid in the Taxonomic_Name column, all other images classified to the subspecies level are labeled as non-hybrid, and the parent species of the one species-level hybrid is labeled as non-hybrid (only one of them is present in the dataset).
  • filename: Unique filename assigned to image at download (<X>_<Image_name>.<jpg/tif/CR2>) using scripts/download_jiggins_subset.py.
  • filepath: Filepath of the downloaded image (<image_folder>/<Taxonomic_Name>/<filename>) using scripts/download_jiggins_subset.py.
  • md5: MD5 of the downloaded image. This was used as the measure of uniqueness (at the pixel level) to address duplication of images across Zenodo records.

Note:

  • Jiggins_Zenodo_dorsal_Img_Master.csv also has a column CAM_dupe indicating whether the CAMID has multiple images of the same file type within this subset. Most (11,452) specimens have only one dorsal image per file type.
    • CAMIDs are necessarily duplicated for the images that are of just a dorsal forewing or hindwing, so we label those as single_wing. There are one RAW and one JPG image for each single wing. None of these are Heliconius butterflies.
    • Instances of both dorsal and ventral wings in the same image are labeled both-wings; there are 18 such images, all of Heliconius butterflies.
    • There are multiple JPG images with the same CAMID. Note that this does not necessarily mean these are duplicates of the same images; check the views to confirm. There are also JPG copies provided alongside RAW images. Generally, RAW images will be unique up to View, as there is only one CAMID with two RAW images that aren't of just a single wing.

Dataset Creation

Curation Rationale

The Butterfly Genetics Group has a large collection of butterfly images distributed across 31 Zenodo records. They do not all have the same information, and it is sometimes only provided in the record, but not the metadata. With this collection, we combine the provided information (metadata) into a shared format that is easily ingested into ML pipelines. We also add some other labels of interest (based on identification determined by the Butterfly Genetics Group), and endeavor to remove duplication, noting potential points of duplication and providing some assessment tools to help prevent data leakage. This de-duplication effort reduced the overall dataset to covering only 29 of these records, and it is documented in the deduplication_process directory.

Additionally, these datasets are prepared in a format that allows for easy integration with the Imageomics Institute's Data Dashboard for distribution statistics and easy sampling of images by taxonomic information and view.

Source Data

These images are a subset of the Butterfly Genetics Group's Cambridge butterfly collection. This collection of butterfly images comes from the research group of Chris Jiggins at the University of Cambridge derived from almost 20 years of field studies.

Data is pulled from the Zenodo Records in licenses.json. This file also contains full citation information for all records.

Initial Data Collection and Annotation

These images are of a mix of wild-caught and lab-bred butterflies, classified by the Butterfly Genetics Group.

Personal and Sensitive Information

None

Considerations for Using the Data

Discussion of Biases and Other Known Limitations

  • This dataset is imbalanced. Even among the Heliconius subset, some subspecies are more heavily represented than others.
  • Not all images are labeled with the same detail. There are a mix of valid subspecies and hybrids that are labeled as such, but there are also images of butterflies classified only to the genus or species level, for which such a designation may not be clearly made. These images are limited to classification tasks only down to the level of their label. There are also instances of "Heliconius hybrid", where the parent subspecies are not indicated because it was labeled only as a hybrid of that species.
  • There may be overlap between images in record 2548678 and record 3082688.
  • There are multiple images of the same specimen for many specimens; sometimes this is due to different views (eg., dorsal or ventral side), sometimes it is due to JPG copies of the RAW photos, though it seems it is also sometimes that new photos were taken of the same specimen at a different times.
    • The master files contain only images that were determined to be unique (at the pixel level) through MD5 checksum. This does not guarantee that there are not images that are cropped copies of other photos. For instance, record 2553977 has a number of images with _cut_ in their name, some of which are close-up crops of the butterfly wings (though not all, some are just close-ups of the full butterfly).

Please see the deduplication_process/README.md for more details on the deduplication process and recommendations on how to split this data effectively without data leakage.

Additional Information

Dataset Curators

  • Christopher Lawrence (Princeton University) - ORCID: 0000-0002-3846-5968
  • Chris Jiggins (University of Cambridge) - ORCID: 0000-0002-7809-062X
  • Butterfly Genetics Group (University of Cambridge)
  • Elizabeth G. Campolongo (Imageomics Institute, The Ohio State University) - ORCID: 0000-0003-0846-2413

Licensing Information

The data (images and text) are all licensed under Creative Commons Attribution 4.0 International. Each image and text in this dataset is provided under the least restrictive terms allowed by its licensing requirements as provided to us (i.e, we impose no additional restrictions past those specified by this license on the original source files). Modified images are only restricted by this original license.

Images can be matched to their source record through the zenodo_link or record_number column in the Master CSVs to the url or record_number in the licenses.json file, respectively.

Citation Information

Christopher Lawrence, Chris Jiggins, Butterfly Genetics Group (University of Cambridge), Elizabeth G. Campolongo. (2024). Jiggins Heliconius Collection. Hugging Face. https://huggingface.co/datasets/imageomics/Jiggins_Heliconius_Collection.

If you use this dataset, please cite the original datasets (citations for all 29 included Zenodo records are in jiggins.bib) as well as this curated subset.

Contributions

The Imageomics Institute is funded by the US National Science Foundation's Harnessing the Data Revolution (HDR) program under Award #2118240 (Imageomics: A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning). Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.