Modelpublic

imageomics/butterfly_detection_yolo

sourceHugging Facemitupdated 10mo agoView on Hugging Face

0likes

Model Card

Model Card for butterflydetectionyolo

This model takes in images of butterflies as photographed for museum collections and detects butterfly components (L/R forewings, L/R hindwings and body) as well as color checkers and metadata labels. The detection model described here is used in the repository https://github.com/Imageomics/wing-segmentation to detect components and use Meta's Segment-Anything (SAM) model for segmentation of components.

Model Details

yolodetection8mshear10.0scale0.5translate0.1fliplr0.0_best.pt is the butterfly detection model.

The yolo v8 detection model was trained on a dataset of 800 total images from the Heliconius Collection-Cambridge Butterfly, OM_STRI, and Monteiro datasets. The model uses the pretrained yolov8m.pt model.

Model Description

The model is responsible for taking an input image (RGB) and generating bounding boxes for all classes below that are found in the image. Data augmentations applied during training include shear (10.0), scale (0.5), and translate (0.1). The model was trained for 50 epochs with an image size of 256. Note that despite defining an image size of 256, the normalized masks predicted by yolo can be rescaled to the original image size.

Segmentation Classes

[pixel class] corresponding category

—[0] background
—[1] right_forewing
—[2] left_forewing
—[3] right_hindwing
—[4] left_hindwing
—[5] ruler
—[6] white_balance
—[7] label
—[8] color_card
—[9] body

Details

model.train(data=YAML, imgsz=256, epochs=50, batch=16, device=DEVICE, optimizer='auto', verbose=True, val=True, shear=10.0, scale=0.5, translate=0.1, fliplr = 0.0 )

Metrics

Class Images Instances Box(P R mAP50 mAP50-95) all 64 358 0.979 0.887 0.919 0.877 background 64 3 1 0 0.315 0.169 rightforewing 64 58 0.995 0.983 0.986 0.977 leftforewing 64 51 0.975 1 0.985 0.982 righthindwing 64 59 0.997 0.966 0.993 0.977 lefthindwing 64 50 0.975 1 0.993 0.98 ruler 64 31 0.951 1 0.995 0.952 whitebalance 64 18 0.984 1 0.995 0.995 label 64 50 0.996 1 0.995 0.935 colorcard 64 24 0.988 1 0.995 0.992 body 64 14 0.928 0.921 0.939 0.815

Developed by: Michelle Ramirez

How to Get Started with the Model

To view applications of how to load in the model file and predict masks on images, please refer to this github repository

Citation

BibTeX:

@software{Ramirez_Lepidoptera_Wing_Segmentation_2024,
  author = {Ramirez, Michelle},
  doi = {10.5281/zenodo.10869579},
  month = mar,
  title = {{Lepidoptera Wing Segmentation}},
  url = {https://github.com/Imageomics/wing-segmentation},
  version = {1.0.0},
  year = {2024}
}

APA:

Ramirez, M. (2024). Lepidoptera Wing Segmentation (Version 1.0.0) [Computer software]. https://doi.org/10.5281/zenodo.10869579

Acknowledgements

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.