Dataset Card for Curated Gold Standard Hoyal Cuthill Dataset

Dataset Description

Dorsal full body images of subspecies of Heliconius erato and Heliconius melpomene (18 subspecies total). There are 960 images with 320 specimens (3 images of each specimen: Original/ Bird transformed/ Butterfly transformed) The original images are low-resolution RGB photographs (photographs were "cropped and resized to a height of 64 pixels (maintaining the original image aspect ratio and padded to 140 pixels wide)"(Hoyal Cuthill et al., 2019)). These low-resolution images were then transformed using AcuityView with estimates of acuity from AcuityView 2.0 and (Land, 1997). Users should know that since (Land, 1997), more recent estimates of insect visual acuity have been created.

This data represents a subset of images processed from Hoyal Cuthill et al. dataset available at doi:10.5061/dryad.2hp1978. Their original dataset also includes ventral images.

Note: dorsal_images_cuthill contains processed dorsal images from the original Hoyal Cuthill dataset (all 1,234 specimens).

Homepage:
Repository: Butterfly-mimicry contains research done using this dataset.
Paper: Imageomics Approach to Understanding Visual Biological Trait Similarities using Butterfly Mimicry as a Model System
Leaderboard:
Point of Contact:

Dataset Summary

Supported Tasks and Leaderboards

Heliconius erato and Heliconius melpomene subspecies identification (image classification), with variable settings for acuity of the observer (bird, butterfly, or human/other).

Languages

English

Dataset Structure

|-- dorsal_images_cuthill   
|		|    
|		|-- 10427965_D_lowres.tif    
|		|    
|		|-- 10427966_D_lowres.tif    
|		|    
|		|	...    
|    
|-- Acuities    
|		|    
|		|-- train_bird    
|		|		|    
|		|		|-- erato_cyrbia    
|		|		|    
|		|		|-- erato_etylus    
|		|		|    
|		|		|	...     
|		|    
|		|-- test_bird    
|		|		|    
|		|		|-- erato_cyrbia    
|		|		|    
|		|		|-- erato_etylus    
|		|		|    
|		|		|	...     
|		|    
|		|-- train_butterfly    
|		|		|    
|		|		|-- erato_cyrbia    
|		|		|    
|		|		|-- erato_etylus    
|		|		|    
|		|		|	...     
|		|    
|		|-- test_butterfly    
|				|    
|				|-- erato_cyrbia    
|				|    
|				|-- erato_etylus     
|				|    
|				|	...     
|    
|-- train    
|		|    
|		|-- erato_cyrbia    
|		|    
|		|-- erato_etylus    
|		|    
|		|	...     
|    
|-- test    
        |     
        |-- erato_cyrbia    
        |    
        |-- erato_etylus    
        |    
        |	...

Data Instances

Type: PNG
Size: 128px x 128px
Background: [210, 210, 210] (gray)
Fit in frame: Most padding is above and below the image, some on the left and right.
Ruler or Scale: None
Color Reference (ColorChecker, white-balance, None): None

Data Fields

In Hoyal_Cuthill_GoldStandard_metadata_cleaned.csv:

NHM_Specimen: Natural History Museum Specimen number
Image_filename: filename of image of specimen
View: whether ventral or dorsal view of specimen (all dorsal)
Species: species of specimen (melpomene or erato)
Subspecies: subspecies of the specimen
Sex: sex of the specimen (male or female)
addit_taxa_info: additional taxonomic information (subspecies)
type_stat: indicates "classical" or "example" specimen of species or subspecies ('ST', 'PT', or 'HT', indicating syntypes, paratypes, or holotypes, respectively). This field is mostly null.
hybrid_stat: hybrid status ('valid subspecies', 'subspecies synonym' or 'unknown' (only 1))
in_reduced: whether or not the specimen was used in the second analysis by Hoyal Cuthill et al. (1 or 0 to indicate yes or no, respectively). This was an effort to remove potential hybrids from their analysis; it does not always match our indication of hybrid status.
locality: where specimen was collected
lat: latitude where specimen was collected
lon: longitude where specimen was collected
speciesdesig: species designation, first initial of species '.' subspecies (eg., 'm. rosina')

Train_Test_Curated_GoldStandard_Hoyal_Cuthill.csv has three additional columns:

Image_filename_png: filename of (png) image of specimen, dorsal_images_cuthill/ + <Image_filename_png> is the filepath for the processed dorsal image
subset: whether this is part of the training or test set (train or test)
filepath: the filepath for the train or test image

Acuity_Curated_GoldStandard_Hoyal_Cuthill.csv also has Image_filename_png and subset, but subset references the bird and butterfly acuity training and test sets. Additionally, it has columns:

bird_filepath: the filepath for the bird acuity version of the image
butterfly_filepath: the filepath for the butterfly acuity version of the image

Data Splits

There are 250 images in each training set and 70 in each test set.

RGB training images are in train folder and testing are in test.

For bird and butterfly acuities, their respective training and test images are in the train_bird (train_butterfly) and test_bird (test_butterfly) folders.

All of these folders are further subdivided by the subspecies. Filepaths to access these images are provided in Train_Test_Curated_GoldStandard_Hoyal_Cuthill.csv and Acuity_Curated_GoldStandard_Hoyal_Cuthill.csv, respectively.

Dataset Creation

Processing steps included:

Hybrid Separation
Label Correction
Removal of subspecies with no mimic pairs
Make background uniform across all images
Make image square via padding

Curation Rationale

This dataset was curated for training a model to classify different species of Heliconius Butterflies and to take into account mimicry between species and acuity of the observer (bird, butterfly, or human/other). The original data (Hoyal Cuthill et al. 2019) had misclassified species/subspecies and some locality/ collection sites were outside the known range of the butterflies. It also contained hybrid and aberrant samples, that had the potential to muddle classification results. To prevent this, the data was further refined by several Heliconius experts to remove hybrid and aberrant samples. Lastly, bird and butterfly acuities were added to provide another level of analysis using AcuityView and estimates of observer acuity.

Source Data

Hoyal Cuthill et al. doi:10.5061/dryad.2hp1978.

Initial Data Collection and Normalization

Photographers: Robyn Crowther and Sophie Ledger, Natural History Museum, London.

Who are the source language producers?

[More Information Needed]

Annotations

Annotation process

The original data has some misclassified species/subspecies, and had multiple hybrid samples. These samples were removed by hand by Owen McMilan, Christopher Lawrence, Jim Mallet, Krzysztof Kozak. Some localities were outside the known range of the butterflies, and were removed using QGIS and known subspecies ranges.

Who are the annotators?

Christopher Lawrence, Jim Mallet, Owen McMilan, and Krzysztof Kozak.

Personal and Sensitive Information

None

Considerations for Using the Data

Social Impact of Dataset

N/A

Discussion of Biases

Biased towards species and subspecies within Heliconius. Focused on Heliconius erato and Heliconius melpomene.

Other Known Limitations

No genetic data available.
Non-uniform distribution of subspecies (imbalanced).

Additional Information

Dataset Curators

Krzysztof Kozak (University of California Berkeley) - ORCID: 0000-0001-8980-3173
Christopher Lawrence (Princeton University) - ORCID: 0000-0002-3846-5968
James Mallet (Harvard University) - ORCID: 0000-0002-3370-0367
Owen McMillan (Smithsonian Tropical Research Institute) - ORCID: 0000-0003-2805-2745
David Carlyn (The Ohio State University) - ORCID: 0000-0002-8323-0359
Mohannad Elhamod (Virginia Tech) - ORCID: 0000-0002-2383-947X

Licensing Information

This work has been marked as dedicated to the public domain by applying the CC0 Public Domain Waiver.

Citation Information

Krzysztof Kozak, Christopher Lawrence, James Mallet, Owen McMillan, David Carlyn, Mohannad Elhamod. (2023), "Curated GoldStandard Hoyal Cuthill", https://huggingface.co/datasets/imageomics/Curated_GoldStandard_Hoyal_Cuthill.

Ramesh Babu, R. (2023). Imageomics Approach to Understanding Visual Biological Trait Similarities using Butterfly Mimicry as a Model System [Master's thesis, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu168198420667979

Please also cite the original dataset from which this was adapted and its accompanying paper:

Hoyal Cuthill, Jennifer F. et al. (2019), Data from: Deep learning on butterfly phenotypes tests evolution’s oldest mathematical model, Dryad, Dataset, https://doi.org/10.5061/dryad.2hp1978.
Hoyal Cuthill, Jennifer F. et al. (2019), Deep learning on butterfly phenotypes tests evolution’s oldest mathematical model, Science Advances, Article-journal, https://doi.org/10.1126/sciadv.aaw4967.

BibTeX

Dataset:

@misc{CGSHC23,
  author = {Krzysztof Kozak and Christopher Lawrence and James Mallet and Owen McMillan and David Carlyn and Mohannad Elhamod},
  title = {Curated GoldStandard Hoyal Cuthill},
  year = {2023},
  url = {https://huggingface.co/datasets/imageomics/Curated_GoldStandard_Hoyal_Cuthill},
  doi = {doi:10.57967/hf/1351},
  publisher = {Hugging Face}
}

Imageomics paper (Part of thesis work):

@masterthesis{ramesh_babu23,
  title    = {Imageomics Approach to Understanding Visual Biological Trait Similarities using Butterfly Mimicry as a Model System},
  author   = {Reshma Ramesh Babu},
  year     =  2023,
  month    = {May},
  note     = {Available at \url{http://rave.ohiolink.edu/etdc/view?acc_num=osu168198420667979}},
  school   = {The Ohio State University},
  type     = {Master's thesis}
}

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.