DualProtoSeg: Simple and Efficient Design with Text- and Image-Guided Prototype Learning for Weakly Supervised Histopathology Image Segmentation

Quick setup

• 1. Download the BCSS dataset and place it under the project data/ folder.
  • Example layout (common):
    • data/train/
    • data/val/
    • data/test/
  • Adjust dataset paths in the config if your layout differs.
• 2. Install CONCH (project-specific):
  • If a pip package is available, you can try:

# inside a venv (recommended)
python3 -m venv .venv
. .venv/bin/activate
pip install --upgrade pip
pip install open_clip_torch

• Or install the CONCH repo (if provided by your team / HF):

# example - replace with the actual CONCH repo URL if needed
git clone https://github.com/MahmoodLab/conch.git
cd conch
pip install -e .

• 3. Install Python dependencies from requirements.txt:

# activate your virtualenv first
. .venv/bin/activate
pip install -r requirements.txt

Note: For torch, pick the wheel that matches your CUDA version. Example (change cu121 to your CUDA version):

pip install --index-url https://download.pytorch.org/whl/cu121 torch torchvision

• 4. Run training with the Makefile

# run with defaults: CONFIG=config.yaml, GPU=0
make run

# or explicitly
make run CONFIG=config.yaml GPU=0

The script will default to saving outputs under a runs/ folder if the config path has no directory component.

Notes & tips

• If your config.yaml points to a CONCH checkpoint (e.g., clip.checkpoint_path), ensure that file path exists.
• To resume training from a checkpoint, pass --resume path/to/checkpoint.pth to train.py (edit Makefile or run python3 train.py ...).

Acknowledgment

Parts of the utils/ folder were inspired by the excellent PBIP implementation by
QingchenTang.
We sincerely thank the authors for making their code publicly available.