TIACentre/Virchow2_Distilled_Segmentation_Models

Model Card

This repository contains pretrained weights of tissue segmentation models distilled from Virchow2, as described in the paper: Computationally Efficient Pathology Segmentation using Knowledge Distillation from Foundation Models

Model Details

PUMA Model

• Developed by: Tissue Image Analytics (TIA) Centre, University of Warwick, United Kingdom.
• Model type: segformerb5
• License: CC BY-NC-SA 4.0
• Model Stats:
  • Params(M): 84.60M
  • Image Size: 224 x 224
  • Resolution: 1.14 mpp
• Tissue Type:
  • Background
  • Stroma
  • Blood Vessels
  • Tumour
  • Epidermis
  • Necrosis
• Training Datasets:
  • PUMA (Supervised)
  • TCGA-SKCM (Unsupervised)
• Pretrained weights:
  • segformerb5_distilled_PUMA_1.pth
  • segformerb5_distilled_PUMA_2.pth
  • segformerb5_distilled_PUMA_5.pth

IGNITE Model

• Developed by: Tissue Image Analytics (TIA) Centre, University of Warwick, United Kingdom.
• Model type: segformerb5
• License: CC BY-NC-SA 4.0
• Model Stats:
  • Params(M): 84.60M
  • Image Size: 448 x 448
  • Resolution: 0.5 mpp
• Tissue Type:
  • Unknown
  • Background
  • Tumour
  • Stroma
  • Necrosis
  • Macrophages
  • Rest
• Training Datasets:
  • IGNITE (Supervised)
  • TCGA-ACC (Unsupervised)
  • TCGA-KIRC (Unsupervised)
  • TCGA-KIRP (Unsupervised)
  • TCGA-KICH (Unsupervised)
  • TCGA-GBM (Unsupervised)
  • TCGA-LIHC (Unsupervised)
  • TCGA-SARC (Unsupervised)
  • CPATC-LUAD (Unsupervised)
  • CPTAC-LSCC (Unsupervised)
• Pretrained weights:
  • segformerb5_distilled_IGNITE_1.pth

BEETLE Model

• Developed by: Tissue Image Analytics (TIA) Centre, University of Warwick, United Kingdom.
• Model type: segformerb5
• License: CC BY-NC-SA 4.0
• Model Stats:
  • Params(M): 84.60M
  • Image Size: 448 x 448
  • Resolution: 1.0 mpp
• Tissue Type:
  • Unknown
  • Other
  • Non_invasive_epithelium
  • Invasive_epithelium
  • Necrosis
• Training Datasets:
  • BEETLE (Supervised)
  • TCGA-BRCA (Unsupervised)
• Pretrained weights:
  • segformerb5_distilled_BETTLE_1.pth

Citation

If you use this model, please cite:

BibTeX:

@misc{lv2026computationallyefficientpathologysegmentation,
title={Computationally Efficient Pathology Segmentation using Knowledge Distillation from Foundation Models}, 
author={Jiaqi Lv and Yijie Zhu and Saki Okada and Maxwell Stanley Renna and Robert Goldin and Stefan Antonowicz and Shan E Ahmed Raza},
year={2026},
eprint={2609.03947},
archivePrefix={arXiv},
primaryClass={eess.IV},
url={https://arxiv.org/abs/2609.03947}, 
}