# Visual Wetlandbirds Eval

> This benchmark evaluates deep learning models on fine-grained bird species classification and spatio-temporal behavior recognition in ecological video footage. It probes the model's ability to localize birds, identify their species, and classify their actions across video frames in real-world wetland environments. Use when the user wants to benchmark on Visual WetlandBirds Dataset, or asks about evaluating this task. Reports accuracy.

- Skill: `qhjqhj00/visual-wetlandbirds-eval` (Agent Skill)
- Install (CLI): `npx skillmds add qhjqhj00/visual-wetlandbirds-eval`
- Raw SKILL.md: https://api.skillmd.com/api/skills/qhjqhj00/visual-wetlandbirds-eval/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: qhjqhj00 (https://skillmd.com/u/qhjqhj00)
- Updated: 2026-09-08
- Page: https://skillmd.com/skills/qhjqhj00/visual-wetlandbirds-eval

---


# visual-wetlandbirds-eval

> Visual WetlandBirds Dataset: Bird Species Identification and Behavior Recognition in Videos — Rodriguez-Juan et al. (2025) (arXiv:2501.08931, 2025)

## What this evaluates

This benchmark evaluates deep learning models on fine-grained bird species classification and spatio-temporal behavior recognition in ecological video footage. It probes the model's ability to localize birds, identify their species, and classify their actions across video frames in real-world wetland environments.

## Datasets

- **Visual WetlandBirds Dataset** — total 178; splits: train (-1), val (-1), test (-1); repo https://github.com/3dperceptionlab/Visual-WetlandBirds

## Metrics

- `accuracy` **(primary)** — range: [0, 1]
  - Not specified in provided text. Standard classification accuracy is implied for species and behavior recognition tasks.

## Input / output format

**Input**: Video sequences (resolutions vary: 1920x1080, 1296x720, 1280x720, 960x540, 3840x2160) containing bird footage from Spanish wetlands.

**Output**: Per-frame predictions including bird species identity, bounding box coordinates, and behavior class labels.

## Scoring recipe

```python
def compute_accuracy(predictions, gold):
    correct = sum(1 for p, g in zip(predictions, gold) if p == g)
    return correct / len(gold) if gold else 0.0
```

## Common pitfalls

- Camera traps have fixed perspectives, resulting in short video clips per individual and requiring models to handle limited temporal context.
- Semi-automated annotation relies on YOLOv7 for initial localization, which may introduce bounding box errors that require manual correction.
- Subject tracking across frames is not natively supported by the annotation tool, requiring post-processing scripts to correlate bounding boxes using Euclidean distance.

## Evidence (verbatim from paper)

> To ensure annotation accuracy, the use of annotation tools has been extended, as they provide a user-friendly interface that makes this process easy and accessible to non-technical staff.

## Citation

```bibtex
@misc{rodriguezjuan2025visualwetlandbirds,
  title={Visual WetlandBirds Dataset: Bird Species Identification and Behavior Recognition in Videos},
  author={Rodriguez-Juan et al. (2025)},
  year={2025},
  note={arXiv:2501.08931}
}
```

- arXiv: 2501.08931