0xd8b / wda-net

Weakly-Supervised Domain Adaptive Segmentation of Electron Microscopy

Weakly-Supervised Cross-Domain Segmentation of Electron Microscopy with Sparse Point Annotation
for Adapting Semantic Segmentation

This repository contains the official implementation of our paper:

Introduction

Weakly-Supervised Cross-Domain Segmentation of Electron Microscopy with Sparse Point Annotation,
In: IEEE TRANSACTIONS ON BIG DATA,2024
[arXiv]

Installation

Requirements. To reproduce our results, we recommend Python >=3.6, PyTorch >=1.4, CUDA >=10.0. At least one GPU with a minimum of 11GB memory is required for training.

1. Create conda environment:

conda create --name wda-net
source activate wda-net

2. Install PyTorch >=1.4 (see PyTorch instructions). For example,

conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch

3. Install the dependencies:

pip install -r requirements.txt

4. Download data (VNC, EPFL and create symlinks in the ./data folder, as follows:

./data/50%vncdata -> <symlink to VNC>
./data/50%vncdata/train/img
./data/50%vncdata/train/lab

./data/cvlabdata -> <symlink to EPFL>
./data/cvlabdata/train/img
./data/cvlabdata/train/lab
./data/cvlabdata/train/15%_split1
./data/cvlabdata/test/img
./data/cvlabdata/test/lab

Tip: We provide labels with 15% center points in target domain. You can download the data: 15%_split1

Training

0.Run the count model (Source domain: VNC)

python 00_count_main.py

If you would like to skip this step, you can use our pre-trained models: vnc_count.pth, Tip: You can download the file and create symlinks in the ./pretrain_model folder, as follows: ./pretrain_model/vnc_count.pth:

1.Run the pretrained model (Source domain: VNC)

python 00_Full-Supervised.py

If you would like to skip this step, you can use our pre-trained models: vnc_full_supervised.pth, Tip: You can download the file , as follows: ./pretrain_model/vnc_full_supervised.pth:

2.Run the detection model (Target domain: EPFL)

python 01_stage1_vnc.py

3.Run the segmentation model (Target domain: EPFL)

python 02_stage2_vnc.py

Inference

Run single-scale inference from your model

python evalue_segmentation.py

You can use our pre-trained model: inference.pth

Citation

We hope you find our work useful. If you would like to acknowledge it in your project, please use the following citation:

@article{qiu2024weakly,
  title={Weakly-Supervised Cross-Domain Segmentation of Electron Microscopy with Sparse Point Annotation},
  author={Qiu, Dafei and Xiong, Shan and Yi, Jiajin and Peng, Jialin},
  journal={IEEE Transactions on Big Data},
  year={2024},
  publisher={IEEE}
}