v1.0

This is mothra, release 1.0.
Mothra analyzes images of lepidopterans — mainly butterflies and moths — using deep learning and image processing, returning the following outputs:

• the length of their wings.
• the position of the lepidoptera: right-way up or upside down.
• the gender of the lepidoptera.

For usage instructions and a complete list of measurement results, please check the README. Images used to train the deep learning models used in mothra are available at the repository GL:mothra/mothra-data.

If you find mothra useful, please cite:

Wilson, R. J. et al. (2022). Applying computer vision to digitised natural history collections for climate change research: Temperature-size responses in British butterflies. Methods in Ecology and Evolution, 1–13.

Mothra's logo was designed by Rogério Maroja, and is available under the CC-BY license.

v1.0-rc.2

In this release candidate, we:

• fix a bug in shoulder detection.
• add and refactor tests for functions in pipeline.py and binarization.py.
• change the pipeline to not break/stop if a specific image returns an error.

v1.0-rc.1

In this release candidate, we:

• update fastai version to 2, and refactor all necessary code.
• set U-Net as the main segmentation algorithm, and remove the options for Grabcut and Otsu.
• rewrite the deep learning segmentation model: instead of returning only the lepidopteran, it now recognizes the ruler and labels as well.
• rewrite the classification models: instead of two models, id_position (returning "right way up" or "upside down") and id_gender (when "right way up", returning "male" or "female"), we now have id_gender returning three classes: "upside down", "male" or "female".

v0.3

Changes in this release:

• The README was updated, and examples used in it were moved to their own folder, mothra/examples.
• The data repository containing the models and images used for training was updated from gitlab:alexdesiqueira/mothra-models to gitlab:mothra/mothra-data.
• Both classic (Otsu threshold) and grabcut segmentation options were removed, leaving only U-nets.
  • API changes: the pipeline does not accept the options -g (grabcut segmentation) and -u (U-net segmentation) anymore. The pipeline operation is equivalent of using -u in mothra-v0.2.

v0.2

Several additions to the pipeline.

• Option of using U-nets for butterfly segmentation
• Recognition of position (right-way up, upside down), using ResNet-34
• Recognition of gender (female, male) when right-way up, using ResNet-50
• Measurement of shoulder distance

Also, improvements under the hood. Now, the pipeline:

• Sets the wing tip as the farthest point to the center of the butterfly, instead of the closest point to the corner of the image
• Doesn't overwrite results.txt files previously saved in the same folder
• Allows batches of image filenames to be input via text file

v0.1

This first release of the butterfly analysis library provides a proof-of-concept implementation of a fully automated pipeline that can segment a butterfly and measure several of its attributes.