Skip to content
SaamieVinckenPublic

About

A Vision Transformer based approach to Strong Lens classification

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

57 Commits

Folders and files

Repository files navigation

AstroVink

Version: AstroVink-Q1

AstroVink is a Vision Transformer (ViT-S/14, DINOv2 backbone) model fine-tuned for automated strong gravitational lens detection on Euclid Space Telescope images.

The model architecture and training setup are described in the paper
“Euclid Quick Data Release (Q1): AstroVink: A vision transformer approach to find strong gravitational lens systems.”
(Vincken, S. H. et al., in prep., Astronomy & Astrophysics), Arxiv link: https://arxiv.org/abs/2604.21977.

Installation

Use the package manager pip to install AstroVink.

pip install -e .

Repository structure

astro_vink/ – model, inference, transforms, metrics, training and utilities
examples/ – example scripts (no data included)
weights/ – optional folder for local model checkpoints

Get the AstroVink weights

The AstroVink-Q1 model weights are available on Hugging Face:

https://huggingface.co/SaamieVincken/AstroVink-Q1

These weights are released under the MIT License, consistent with this repository.

Usage

After installation, you can run inference on an image using a fine-tuned model checkpoint.

Inference example

A ready to run example script is provided at examples/run_inference.py

Which can be executed directly from the project root:

python examples/run_inference.py

This script demonstrates how to:

  • Load the model and weights
  • Run inference on an image
  • Print the lens and non-lens probabilities

Training examples

Example scripts are also provided for training and fine-tuning:

examples/run_training.py trains the AstroVink (base) model from scratch.

examples/run_retraining.py fine-tunes the model on new data.

These can be executed directly from the project root:

python examples/run_training.py
python examples/run_retraining.py

Note: This repository does not include any model weights or images. To run the example, you must provide your own image file and specify the path to the weights.

Data availability

The data AstroVink-Q1 is applied to is available on Zenodo as the Euclid Quick Data Release (Q1): The Strong Lensing Discovery Engine (SLDE).

Citation

If you use AstroVink-Q1 or any part of this repository in academic work, please cite:

Vincken, S. H. (2026).
Euclid Quick Data Release (Q1). AstroVink: A vision transformer approach to find strong gravitational lens systems.
Astronomy & Astrophysics (in preparation). https://arxiv.org/abs/2604.21977.

A complete citation record is available in the CITATION.cff file.

License

This project is released under the MIT License.
See the LICENSE file for the full license text.
The license applies only to the code in this repository.
All Euclid data remain proprietary and are not included under this license.

Contact

For questions or collaborations related to AstroVink, please contact:

Saamie Helena Vincken

Email: saamievincken@protonmail.com

GitHub: @saamievincken

About

A Vision Transformer based approach to Strong Lens classification

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages