davanstrien/ena24-detection

---
license: cdla-permissive-1.0
task_categories:
- object-detection
tags:
- wildlife
- camera-trap
- conservation
- object-detection
- bounding-boxes
- animals
- ecology
size_categories:
- 1K<n<10K
---

# ENA24 - Eastern North America Camera Trap Object Detection

Camera trap images with bounding box annotations for 23 wildlife species from Eastern North America.

## Dataset Description

This dataset contains 8,789 camera trap images with 9,772 bounding box annotations across 23 species. Images are 1920x1080 resolution. Annotations are in COCO format (xywh bounding boxes).

Originally collected by the University of Missouri and distributed via [LILA BC](https://lila.science/datasets/ena24detection/) (Labeled Information Library of Alexandria: Biology and Conservation).

## Species (23 categories)

| Species | Count | | Species | Count |
|---|---|---|---|---|
| American Crow | 1,278 | | Eastern Cottontail | 340 |
| American Black Bear | 959 | | Bobcat | 333 |
| Chicken | 732 | | Eastern Gray Squirrel | 319 |
| Dog | 726 | | Eastern Chipmunk | 311 |
| Virginia Opossum | 725 | | Striped Skunk | 297 |
| Domestic Cat | 482 | | Northern Raccoon | 292 |
| Grey Fox | 438 | | Bird | 225 |
| Wild Turkey | 427 | | Woodchuck | 206 |
| Red Fox | 413 | | Vehicle | 138 |
| White-Tailed Deer | 381 | | Horse | 63 |
| Coyote | 344 | | Human | 0 |
| Eastern Fox Squirrel | 343 | | | |

## Dataset Structure

Each example contains:
- `image`: Camera trap photograph (PIL Image)
- `image_id`: Unique image identifier
- `width` / `height`: Image dimensions
- `objects`: Object detection annotations
- `bbox`: List of bounding boxes in COCO format `[x, y, width, height]`
- `category`: List of species labels (`ClassLabel`)
- `area`: List of bounding box areas

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("davanstrien/ena24-detection")
```

### Train an object detector

This dataset is compatible with the [Hugging Face vision-trainer skill](https://github.com/huggingface/skills/tree/main/skills/hugging-face-vision-trainer) for fine-tuning object detection models:

```bash
hf jobs uv run --flavor l4x1 \
https://huggingface.co/datasets/hugging-face-vision-trainer/scripts/raw/main/object_detection_training.py \
--dataset_name davanstrien/ena24-detection \
--model_name_or_path facebook/detr-resnet-50 \
--output_dir wildlife-detector \
--do_train --do_eval \
--push_to_hub
```

## Source and License

- **Source**: [LILA BC - ENA24 Detection](https://lila.science/datasets/ena24detection/)
- **License**: [Community Data License Agreement - Permissive 1.0](https://cdla.dev/permissive-1-0/)
- **Original contributor**: University of Missouri
- **COCO JSON converted by**: Vardhan Duvvuri
- **Citation**:

```bibtex
@misc{ena24,
title={ENA24 Detection Dataset},
author={University of Missouri},
year={2016},
url={https://lila.science/datasets/ena24detection/}
}
```

## Acknowledgments

This dataset was originally hosted on [LILA BC](https://lila.science/), a repository for labeled information supporting conservation biology research. Redistributed here under the CDLA Permissive 1.0 license.