This repository hosts pre-trained models for beetle localization (object detection) and elytra coordinate prediction (keypoint regression), enabling automated morphological trait measurement from beetle images.

These models accompany the paper:

Yoo, H., Somasundaram, D., and Oh, H. (2025). Efficient CNN-Based System for Automated Beetle Elytra Coordinates Prediction. VISAPP 2025, VISIGRAPP. DOI: 10.5220/0013264600003912

Model Overview

Two types of models are provided (see GitHub repository for full usage details):

Object Detection — YOLOv8 models trained to detect and localize individual beetles in group images
Regression — CNN-based models trained to predict elytra keypoint coordinates for morphological trait analysis (length and width)

Dataset

Models were trained on the yoohj0416/predictbeetle dataset, a re-created version of the 2018-NEON-beetles dataset augmented with:

Bounding box annotations (extracted using SAM + contour detection)
Manually refined elytra keypoint coordinates

Model Performance

Object Detection (YOLOv8)

Model	AP50	mAP
YOLOv8n	0.968	0.800
YOLOv8s	0.970	0.805
YOLOv8m	0.971	0.804

Elytra Coordinate Regression

Backbone	MSE	Points Difference (cm)	Params
ResNet50	1.941E-03	0.128	23.5M
ResNet101	1.971E-03	0.132	42.5M
MobileNetV3-Large	1.952E-03	0.118	4.2M
EfficientNetV2-S	1.870E-03	0.110	20.2M
EfficientNetV2-M	1.756E-03	0.110	52.9M

Intended Uses

Automated detection and localization of beetles in high-resolution group images
Keypoint regression for elytra length and width measurement
Ecological and biodiversity research requiring large-scale morphological analysis

Citation

If you use these models, please cite:

bibtex

@inproceedings{Yoo2025EfficientCNN,
  author    = {Yoo, H. and Somasundaram, D. and Oh, H.},
  title     = {{Efficient CNN-Based System for Automated Beetle Elytra Coordinates Prediction}},
  booktitle = {Proceedings of the 20th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications ({VISIGRAPP} 2025) - Volume 2: {VISAPP}},
  pages     = {934--941},
  year      = {2025},
  publisher = {{SCITEPRESS – Science and Technology Publications, Lda.}},
  doi       = {10.5220/0013264600003912},
  isbn      = {978-989-758-728-3},
  issn      = {2184-4321},
  series    = {{VISIGRAPP} 2025}
}

Please also cite the original dataset:

bibtex

@misc{Fluck2018_NEON_Beetle,
  author    = {Isadora E. Fluck and Benjamin Baiser and Riley Wolcheski and Isha Chinniah and Sydne Record},
  title     = {2018 {NEON} Ethanol-preserved Ground Beetles (Revision 7b3731d)},
  year      = {2024},
  url       = {https://huggingface.co/datasets/imageomics/2018-NEON-beetles},
  doi       = {10.57967/hf/5252},
  publisher = {Hugging Face}
}

Specimens:

bibtex

@misc{Portal2022-ho,
  title     = "{NEON} biorepository Carabid collection (trap sorting)",
  author    = "Portal, Neon Biorepository",
  publisher = "National Ecological Observatory Network",
  doi = {https://doi.org/10.15468/mjtykf},
  note = {Accessed in 2022}
}

bibtex

@misc{Portal2022-qu,
  title     = "{NEON} Biorepository Carabid Collection (Archive Pooling)",
  author    = "Portal, Neon Biorepository",
  publisher = "National Ecological Observatory Network",
  doi = {https://doi.org/10.15468/xicbza},
  note = {Accessed in 2022}
}

Acknowledgements

This work was supported by the NSF OAC 2118240 Imageomics Institute award and was initiated at Beetlepalooza 2024.

This material is based in part upon work supported by the U.S. National Ecological Observatory Network (NEON), a program sponsored by the U.S. National Science Foundation (NSF) and operated under cooperative agreement by Battelle. Specimen data used in this project were collected as part of the NEON Program.

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.