SPROUT Preprocessed Datasets

This repository provides the preprocessed datasets used in SPROUT:

Supervise Less, See More: Training-free Nuclear Instance Segmentation with Prototype-Guided Prompting ICML 2026

Official code repository: https://github.com/Y-Research-SBU/SPROUT

Datasets

The repository contains preprocessed versions of the following publicly available nuclear segmentation datasets:

• MoNuSeg
• CPM17
• TNBC

The datasets have been reorganized and preprocessed to match the data format expected by the SPROUT codebase.

Directory Structure

The dataset follows the structure used by SPROUT:

MoNuSeg/
├── train/
│   ├── annotations/
│   ├── images/
│   ├── instances/
│   ├── semantics/
│   ├── png/
│   └── train.txt
└── test/
    ├── annotations/
    ├── images/
    ├── instances/
    ├── semantics/
    ├── png/
    └── test.txt

cpm17/
...

TNBC/
...

Usage

Download or clone the dataset and place the dataset folders under the data/ directory of the SPROUT repository:

SPROUT/
└── data/
    ├── MoNuSeg/
    ├── cpm17/
    └── TNBC/

Then follow the instructions in the official SPROUT repository.

Data Source

We do not claim ownership of the original datasets. MoNuSeg, CPM17, and TNBC were released by their respective authors. This repository only provides reorganized/preprocessed versions of these datasets for reproducibility and convenient use with SPROUT.

SPROUT Citation

If you use these preprocessed datasets together with SPROUT, please consider citing:

@inproceedings{zhang2026sprout,
  title={Supervise Less, See More: Training-free Nuclear Instance Segmentation with Prototype-Guided Prompting},
  author={Wen Zhang and Qin Ren and Wenjing Liu and Haibin Ling and Chenyu You},
  booktitle={International Conference on Machine Learning},
  year={2026}
}

License

The preprocessing and organization provided in this repository are intended to support reproducibility of the SPROUT project. The original datasets remain subject to their respective licenses and terms of use.