academic-datasets/Visual-WetlandBirds-Dataset

Dataset Card for Visual WetlandBirds Dataset

The Visual WetlandBirds Dataset is a fine-grained spatio-temporal dataset specifically designed for bird behavior detection and species classification. This version has been converted to work well with the Hugging Face Hub, with the original dataset available at Zenodo.

The dataset was introduced in the paper Visual WetlandBirds Dataset: Bird Species Identification and Behavior Recognition in Videos.

Dataset Details

Dataset Description

The Visual WetlandBirds Dataset contains videos and annotations for bird behavior detection and species classification. The data was collected within Alicante wetlands, specifically within the wetlands of La Mata Natural Park and El Hondo Natural Park (southeastern Spain), as part of the CHAN-TWIN project.

The dataset identifies 13 different bird species, including White Wagtail, Glossy Ibis, Squacco Heron, Black-winged Stilt, Yellow-legged Gull, Common Gallinule, Black-headed Gull, Eurasian Coot, Little Ringed Plover, Eurasian Moorhen, Eurasian Magpie, Gadwall, Mallard, and Northern Shoveler.

• Curated by: University of Alicante researchers (Rodriguez-Juan, Javier; Ortiz-Perez, David; Benavent-Lledo, Manuel; Mulero-Pérez, David; Ruiz-Ponce, Pablo; Orihuela-Torres, Adrian; Garcia-Rodriguez, Jose; Sebastián-González, Esther)
• Language(s): English (dataset documentation)
• License: Creative Commons Attribution 4.0 International

Dataset Sources

• Repository: https://github.com/3dperceptionlab/Visual-WetlandBirds
• Original Dataset: https://zenodo.org/records/14639444

Uses

Direct Use

This dataset is designed for:

• Bird species identification in video content
• Bird behavior recognition and classification
• Training and evaluation of computer vision and deep learning models on fine-grained spatio-temporal tasks
• Research in ecological monitoring and wildlife conservation

Load the dataset

You can load the dataset using the datasets library as follows:

from datasets import load_dataset

ds = load_dataset("academic-datasets/Visual-WetlandBirds-Dataset")

To work with the video data, you must ensure you have torchvision and av installed. For more information, see the docs.

An example row looks like this:

{'video': <torchvision.io.video_reader.VideoReader at 0x7d5648b059d0>,
 'frame': 0,
 'x_max': 364.62,
 'y_max': 191.11,
 'x_min': 574.08,
 'y_min': 394.75,
 'behavior_id': 1,
 'behavior': 'Preening',
 'bird_id': 0,
 'species_id': 0,
 'species': 'White Wagtail'}

Here is an example from the dataset with the bounding box displayed in the image:

Dataset Creation

Curation Rationale

This dataset was collected to contribute to the CHAN-TWIN project, with a focus on bird behavior detection and species classification in wetland environments.

Source Data

Data Collection and Processing

The data was collected in Alicante wetlands, specifically within the wetlands of La Mata Natural Park and El Hondo Natural Park in southeastern Spain. The videos capture various bird species in their natural habitat exhibiting different behaviors.

Who are the source data producers?

The data was collected and processed by researchers from the University of Alicante, including Rodriguez-Juan, Javier; Ortiz-Perez, David; Benavent-Lledo, Manuel; Mulero-Pérez, David; Ruiz-Ponce, Pablo; Orihuela-Torres, Adrian; Garcia-Rodriguez, Jose; and Sebastián-González, Esther.

Annotations

Annotation process

The dataset includes annotations for bird species identification and behavior recognition. Each frame in the videos is annotated with bounding boxes indicating the location of birds, the species ID, and the behavior ID.

Bias, Risks, and Limitations

• The dataset is limited to 13 bird species found in specific wetland environments in southeastern Spain
• The behaviors and species distributions may not generalize to other environments or regions
• Video quality and lighting conditions may vary across the dataset
• The dataset may contain some annotation errors or inconsistencies

Recommendations

Users should be aware that models trained on this dataset may not generalize well to different geographical regions, environments, or bird species not represented in the dataset. Additional data augmentation or transfer learning may be necessary for broader applications.

Citation

@misc{rodriguez2025wetlandbirds,
title={Visual WetlandBirds Dataset: Bird Species Identification and Behaviour Recognition in Videos},
author={Rodriguez-Juan, Javier and Ortiz-Perez, David and Benavent-Lledo, Manuel and Mulero-Pérez, David and Ruiz-Ponce, Pablo and Orihuela-Torres, Adrian and Garcia-Rodriguez, Jose and Sebastián-González, Esther},
month={dec},
year=2024,
publisher={Zenodo},
doi={10.5281/zenodo.14355257},
url={https://doi.org/10.5281/zenodo.14355257}
}