README.md

metadata

license: apache-2.0
tags:
  - vision-transformer
  - image-classification
  - pytorch
  - timm
  - mlp-mixer
  - gravitational-lensing
  - strong-lensing
  - astronomy
  - astrophysics
datasets:
  - parlange/gravit-j24
metrics:
  - accuracy
  - auc
  - f1
paper:
  - title: 'GraViT: A Gravitational Lens Discovery Toolkit with Vision Transformers'
    url: https://arxiv.org/abs/2509.00226
    authors: Parlange et al.
model-index:
  - name: MLP-Mixer-b3
    results:
      - task:
          type: image-classification
          name: Strong Gravitational Lens Discovery
        dataset:
          type: common-test-sample
          name: Common Test Sample (More et al. 2024)
        metrics:
          - type: accuracy
            value: 0.8118
            name: Average Accuracy
          - type: auc
            value: 0.746
            name: Average AUC-ROC
          - type: f1
            value: 0.4426
            name: Average F1-Score

🌌 mlp-mixer-gravit-b3

🔭 This model is part of GraViT: Transfer Learning with Vision Transformers and MLP-Mixer for Strong Gravitational Lens Discovery

🔗 GitHub Repository: https://github.com/parlange/gravit

🛰️ Model Details

• 🤖 Model Type: MLP-Mixer
• 🧪 Experiment: B3 - J24-all-blocks
• 🌌 Dataset: J24
• 🪐 Fine-tuning Strategy: all-blocks

💻 Quick Start

import torch
import timm

# Load the model directly from the Hub
model = timm.create_model(
    'hf-hub:parlange/mlp-mixer-gravit-b3',
    pretrained=True
)
model.eval()

# Example inference
dummy_input = torch.randn(1, 3, 224, 224)
with torch.no_grad():
    output = model(dummy_input)
    predictions = torch.softmax(output, dim=1)
print(f"Lens probability: {predictions[0][1]:.4f}")

⚡️ Training Configuration

Training Dataset: J24 (Jaelani et al. 2024)
Fine-tuning Strategy: all-blocks

🔧 Parameter	📝 Value
Batch Size	192
Learning Rate	AdamW with ReduceLROnPlateau
Epochs	100
Patience	10
Optimizer	AdamW
Scheduler	ReduceLROnPlateau
Image Size	224x224
Fine Tune Mode	all_blocks
Stochastic Depth Probability	0.1

📈 Training Curves

🏁 Final Epoch Training Metrics

Metric	Training	Validation
📉 Loss	0.0373	0.0536
🎯 Accuracy	0.9863	0.9861
📊 AUC-ROC	0.9989	0.9990
⚖️ F1 Score	0.9863	0.9860

☑️ Evaluation Results

ROC Curves and Confusion Matrices

Performance across all test datasets (a through l) in the Common Test Sample (More et al. 2024):

📋 Performance Summary

Average performance across 12 test datasets from the Common Test Sample (More et al. 2024):

Metric	Value
🎯 Average Accuracy	0.8118
📈 Average AUC-ROC	0.7460
⚖️ Average F1-Score	0.4426

📘 Citation

If you use this model in your research, please cite:

@misc{parlange2025gravit,
      title={GraViT: Transfer Learning with Vision Transformers and MLP-Mixer for Strong Gravitational Lens Discovery}, 
      author={René Parlange and Juan C. Cuevas-Tello and Octavio Valenzuela and Omar de J. Cabrera-Rosas and Tomás Verdugo and Anupreeta More and Anton T. Jaelani},
      year={2025},
      eprint={2509.00226},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2509.00226}, 
}

---

Model Card Contact

For questions about this model, please contact the author through: https://github.com/parlange/