README.md

---
license: cc-by-sa-4.0
tags:
  - computer-vision
  - object-detection
  - keypoint-detection
  - regression
  - biology
  - ecology
  - pytorch
datasets:
  - imageomics/2018-NEON-beetles
  - yoohj0416/predictbeetle
pipeline_tag: image-to-image
model_description: "CNN-based models for automated beetle localization (YOLOv8) and elytra keypoint coordinate prediction (ResNet, MobileNet, EfficientNet), enabling large-scale morphological trait measurement from beetle images."
---

# Efficient CNN-Based System for Automated Beetle Elytra Coordinates Prediction

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](https://doi.org/10.5220/0013264600003912)

---

## Model Overview

Two types of models are provided (see [GitHub repository](https://github.com/yoohj0416/predictbeetle) 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](https://huggingface.co/datasets/yoohj0416/predictbeetle) dataset, a re-created version of the [2018-NEON-beetles](https://huggingface.co/datasets/imageomics/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](https://github.com/Imageomics/BeetlePalooza-2024).

This material is based in part upon work supported by the [U.S. National Ecological Observatory Network (NEON)](https://www.neonscience.org/), a program sponsored by the [U.S. National Science Foundation (NSF)](https://www.nsf.gov/) and operated under cooperative agreement by [Battelle](https://www.battelle.org/). 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.