---
dataset_info:
  features:
  - name: filename
    dtype: string
  - name: label
    dtype: string
  - name: image
    dtype: image
  splits:
  - name: raw_images
    num_bytes: 34945123504.804
    num_examples: 3098
  - name: cropped_body
    num_bytes: 2060540969.592
    num_examples: 3098
  - name: cropped_head
    num_bytes: 737914072.544
    num_examples: 3098
  - name: segmented_body
    num_bytes: 1265192956.906
    num_examples: 3098
  download_size: 38280262700
  dataset_size: 39008771503.846
configs:
- config_name: default
  data_files:
  - split: raw_images
    path: data/raw_images-*
  - split: cropped_body
    path: data/cropped_body-*
  - split: cropped_head
    path: data/cropped_head-*
  - split: segmented_body
    path: data/segmented_body-*
task_categories:
- image-classification
- image-segmentation
size_categories:
- 1K<n<10K
license: cc-by-4.0
pretty_name: jaguar_identification
language:
- en
tags:
- biology
---

# Jaguar Re-identification Dataset

This dataset contains images of jaguars from the Porto Jofre region in the Pantanal National Park, Brazil. It was curated for the purpose of developing and evaluating deep learning models for individual jaguar identification for population tracking.

![](https://github.com/andandandand/practical-computer-vision/blob/main/images/jaguars_fo_2.png?raw=true)

## Dataset Description

The [Jaguar Identification Project](https://www.jaguaridproject.com/) aims to track jaguar movements, health, and demographics. This contributes valuable data to conservation strategies, especially considering jaguars are classified as Near Threatened by the [IUCN](https://iucn.org/) due to habitat loss, poaching, and human-wildlife conflict.

The core idea is to use photos taken by citizen scientists, to distinguish individual jaguars (e.g., identifying "Medrosa" vs. "Patricia").

**Origin of Data:**
The images are primarily sourced from Porto Jofre, a wildlife reserve located in the Pantanal National Park, which is home to one of the largest wild jaguar populations and is a popular destination for ecotourism and wildlife observation.

**Goal of the Dataset:**
The primary goal is to facilitate the development of models that can distinguish different individual jaguars from images. This supports:
*   Population tracking and demographic studies.
*   Citizen science efforts
*   Conservation strategies by providing data on jaguar movements and health.

## Dataset Details

**Data Collection and Preprocessing:**
The dataset was prepared through a pipeline involving:
*   Collection of raw images (initially around 40GB, ~4300 images).
*   **Cropping and Segmentation:**
    *   Cropped body images (using Grounding DINO).
    *   Cropped face images.
    *   Segmented body images (using SAM - Segment Anything Model).
    *   Segmented face images.

The dataset includes images of various individual jaguars, such as "Ousado," "Medrosa," and "Patricia."

## Motivation

Protecting endangered and near-threatened species like jaguars starts with understanding their populations and the threats they face, such as wildfires, deforestation, illegal trade, climate change, and human-wildlife conflicts. 

## Project Goals and Model Development

The dataset is used to:
*   Contribute to citizen science efforts at the Pantanal Jaguar ID Project.
*   Support research on re-identification models.

## Citation and License

All images were captured by Abigail Martin and contributors to the Jaguar ID Project https://www.jaguaridproject.com. The rights of ownership to these images remain to the Jaguar ID Project.  
We kindly ask you to keep all credits as shown on the original images. You will find these in the `raw_images` split.  

If you use this dataset in your research, please cite the following:

```bibtex
@misc{jaguar_identification_dataset_2025,
  author    = {Antonio Rueda-Toicen and Abigail Martin and Shahabeddin Dayani and Davide Panza and Aleksandra Kudaeva and Gerard de Melo},
  title     = {Jaguars of Pantanal Re-Identification Dataset},
  year      = {2025},
  publisher = {Hugging Face},
  journal   = {Hugging Face Hub},
  howpublished = {\url{https://huggingface.co/datasets/jaguaridentification/jaguars}} 
}

```