language:
- en
library_name: ultralytics
license: gpl-3.0
pipeline_tag: image-classification
base_model: Ultralytics/YOLOv8
metrics:
- f1-score
- mAP50-95
tags:
- trapper
- trapperai
- ecology
- biology
- wildlife
- animal detection
- species classification
🐺 Overview
The TrapperAI model is responsible for the detection and classification of 18 European mammal species with a 95% F1-score and 93% mAP50-95. This model is based on the fine-tuned YOLOv8-m model and can be loaded and utilized directly through the Ultralytics package interface or via the TRAPPER ecosystem (TrapperAI Worker).
The dataset used for model training and evaluation comprised 401,458 camera trap images from Poland, Germany, Sweden, Austria, and Switzerland. The data repository consisted of 5,680 deployments and 2,944 locations.
List of supported species:
- Bird
- Cat
- Chamois
- Dog
- Eurasian Lynx
- Eurasian Red Squirrel
- European Badger
- European Mouflon
- Fallow Deer
- Gray Wolf
- Hare
- Marten
- Moose
- Red Deer
- Red Fox
- Reindeer
- Roe Deer
- Wild Boar
The recommended image resolution for the model is 1024px. The model's performance enables the processing of ~30,000 images in one hour using a single NVIDIA GPU with more than 11 GB of vRAM.
📥 Installation
$ python3 -m venv env
$ source env/bin/activate
$ pip install ultralytics dill ipython # IPython is optional
🚀 Usage
In [1]: from ultralytics import YOLO
In [2]: model = YOLO("TrapperAI-v02.2024-YOLOv8-m.pt")
In [3]: results = model.predict("fox36-Vulpes-vulpes.jpg")
In [4]: len(results) # how many animals were detected
Out[4]: 1
In [5]: results[0].show() # open image viewer with detection and classification results
In [6]: results[0].boxes.conf # return best confidence score for detection and classification results
Out[6]: tensor([0.9558], device='cuda:0')
In [7]: results[0].boxes.cls # return index value for detection and classification results
Out[7]: tensor([14.], device='cuda:0') # Red Fox
If your image contains more than one object (animal), you will need to iterate through the results list to obtain the confidence score and species index value for each detected object.
🏢 Who is using TRAPPER?
- Mammal Research Institute Polish Academy of Sciences;
- Karkonosze National Park;
- Swedish University of Agricultural Sciences;
- Svenska Jägareförbundet;
- Meles Wildbiologie;
- University of Freiburg Wildlife Ecology and Management;
- Bavarian Forest National Park;
- Georg-August-Universität Göttingen;
- KORA - Carnivore Ecology and Wildlife Management;
- and many more individual scientists and ecologies;
💲 Funders and Partners
🤝 Support
Feel free to add a new issue with a respective title and description on the TRAPPER issue tracker. If you already found a solution to your problem, we would be happy to review your pull request.
If you prefer direct contact, please let us know: contact@os-conservation.org
We also have TRAPPER Mailing List and TRAPPER Slack.
📜 License
Read more in TRAPPER License.