PBIP: Prototype-Based Image Prompting for Weakly Supervised Histopathological Image Segmentation

Official PyTorch implementation of the CVPR 2025 paper:

Prototype-Based Image Prompting for Weakly Supervised Histopathological Image Segmentation
CVPR 2025

📄 Paper

---

🔥 Highlights

• 🏆 CVPR 2025 acceptance
• 🎯 Weakly Supervised Learning: Achieves pixel-level segmentation using only image-level labels
• 🧬 Histopathological Focus: Specialized for medical image analysis
• 🚀 Prototype-Based Design: Novel prototype-based image prompting mechanism

🏗️ Model Architecture

<div align="center"> <img src="Figure/model.png" alt="PBIP Model Architecture" width="800"> <p><em>Overview of the PBIP architecture for weakly supervised histopathological image segmentation</em></p> </div>

🛠️ Installation

Requirements

• Python 3.8+
• PyTorch 1.9+
• CUDA 11.0+ (for GPU training)

Environment Setup

Using requirements.txt

# Create virtual environment
conda create -n pbip python=3.8
conda activate pbip

# Install exact dependencies (recommended for reproducibility)
pip install -r requirements.txt

📊 Dataset

This project uses the BCSS (Breast Cancer Semantic Segmentation) dataset with 5 tissue classes:

| Class | Description | Color | |-------|-------------|-------| | TUM | Tumor | 🔴 Red | | STR | Stroma | 🟢 Green | | LYM | Lymphocyte | 🔵 Blue | | NEC | Necrosis | 🟣 Purple | | BACK | Background | ⚪ White |

Data Structure

data/
├── BCSS-WSSS/
│   ├── train/
│   │   └── *.png  # Training images with class labels in filename
│   ├── test/
│   │   ├── img/   # Test images
│   │   └── mask/  # Ground truth masks
│   └── valid/
│       ├── img/   # Validation images
│       └── mask/  # Ground truth masks

🚀 Quick Start

Downloading pre-trained SegFormer here.

Training Stage 1 & Generate CAMs

# Train the PBIP model
python train_stage_1.py --config ./work_dirs/bcss/classification/config.yaml --gpu 0

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

@inproceedings{pbip2025,
  title={Prototype-Based Image Prompting for Weakly Supervised Histopathological Image Segmentation},
  author={Qingchen Tang and Lei Fan and Maurice Pagnucco and Yang Song},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2025}
}

---

⭐ Star this repo if you find it helpful! ⭐