GastroNetDinov2

GastroNet-5M — ViT-B pretrained weights

This repository contains a Vision Transformer (ViT-Base) model that was used in the study:

GastroNet-5M: A Multicenter Dataset for Developing Foundation Models in Gastrointestinal Endoscopy
Gastroenterology (2025)
DOI: 10.1053/j.gastro.2025.07.030
https://www.sciencedirect.com/science/article/pii/S001650852505797X

The model was pretrained on the GastroNet‑5M dataset using dinov2.
Please cite the paper when using this model, the dataset, or the pretrained weights.

---

🧠 Weights

The pretrained model weights are hosted externally and can be downloaded here:

➡️ https://cortex.thetavision.nl/dataset-provider/listing/2/

Download the file (e.g., dinov2.pth) and place it locally or on your device.

---

🚀 Usage (PyTorch + timm)

# pip install timm
import torch
import timm

# Initialize ViT‑B backbone (no classifier head)

model = timm.create_model("timm/vit_base_patch14_dinov2.lvd142m",
                          pretrained=False,
                          num_classes=0,
                          img_size=336,
                          )

# Update this path to where you downloaded the checkpoint
ckpt_path = "dinov2.pth"
state = torch.load(ckpt_path, map_location="cpu")
state_dict = state['teacher']

# Remove 'module.' prefix if present
clean_state = {k.replace("backbone.", ""): v for k, v in state_dict.items()}
msg = model.load_state_dict(clean_state, strict=False)
print(msg)
model.eval()

---

📄 Citation

If you use this model, please cite the study:

Plain citation

Jong MR, Boers TGW, Fockens KN, Jukema JB, Kusters CHJ, Jaspers TJM, van Eijck van Heslinga RAH, Slooter FC, Struyvenberg MR, Bisschops R, van der Putten JA, de With PHN, van der Sommen F, de Groof AJ, Bergman JJGHM; BONS-AI Consortium.
GastroNet‑5M: A Multicenter Dataset for Developing Foundation Models in Gastrointestinal Endoscopy. Gastroenterology. 2025. DOI:10.1053/j.gastro.2025.07.030.

BibTeX

@article{Jong2025GastroNet5M,
  author = {Jong, Martijn R and Boers, Tim G. W. and Fockens, Kiki N and Jukema, Jelmer B and Kusters, Carolus H. J and Jaspers, Tim J. M and van Eijck van Heslinga, Rixta A. H and Slooter, Floor C and Struyvenberg, Maarten R and Bisschops, Raf and van der Putten, Joost A and de With, Peter H. N and van der Sommen, Fons and de Groof, Albert J and Bergman, Jacques J. G. H. M. and Barrett's Oesophagus Imaging for Artificial Intelligence (BONS-AI) Consortium},
  title = {GastroNet-5M: A Multicenter Dataset for Developing Foundation Models in Gastrointestinal Endoscopy},
  journal = {Gastroenterology},
  year = {2025},
  doi = {10.1053/j.gastro.2025.07.030}
}