trellixvulnteam / mmotu_ds2net_8m44

MMOTU_DS²Net

This repo is the implementation of "A Multi-Modality Ovarian Tumor Ultrasound Image Dataset for Unsupervised Cross-Domain Semantic Segmentation". we refer to MMSegmentation and MMGeneration and mix them to implement unsupervised domain adaptation based segmentation (UDA SEG) task. Many thanks to SenseTime and their two excellent repos.

Dataset

Multi-Modality Ovarian Tumor Ultrasound (MMOTU) image dataset consists of two sub-sets with two modalities, which are OTU_2d and OTU_CEUS respectively including 1469 2d ultrasound images and 170 CEUS images. On both of these two sub-sets, we provide pixel-wise semantic annotations and global-wise category annotations. Many thanks to Department of Gynecology and Obstetrics, Beijing Shijitan Hospital, Capital Medical University and their excellent works on collecting and annotating the data.

MMOTU : google drive (move OTU_2d and OTU_3d to data folder. Here, OTU_3d folder indicates OTU_CEUS in paper.)

DS²Net

Install

1. requirements:

   python >= 3.7

   pytorch >= 1.4

   cuda >= 10.0

2. prerequisites: Please refer to MMSegmentation PREREQUISITES; Please don't forget to install mmsegmentation with

   cd MMOTU_DS2Net
   
   pip install -e .
   
   chmod 777 ./tools/dist_train.sh
   
   chmod 777 ./tools/dist_test.sh
   

Training

mit_b5.pth : google drive (Before training Segformer or DS²Net_T, loading ImageNet-pretrained mit_b5.pth is very useful. We provide this pretrained backbone here. The pretrained backbone has already been transformed to fit for our repo.)

Task1: Single-modality semantic segmentation

 cd MMOTU_DS2Net
 
 ./tools/dist_train.sh ./experiments/pspnet_r50-d8_769x769_20k_MMOTU/config/pspnet_r50-d8_769x769_20k_MMOTU.py 2

Task2: UDA semantic segmentation

 cd MMOTU_DS2Net
 
 ./tools/dist_train.sh ./experiments/DS2Net_segformerb5_769x769_40k_MMOTU/config/DS2Net_segformerb5_769x769_40k_MMOTU.py 2

Task3: Single-modality recognition:

Testing

Task1: Single-modality semantic segmentation

 cd MMOTU_DS2Net
 
 ./tools/dist_test.sh ./experiments/pspnet_r50-d8_769x769_20k_MMOTU/config/pspnet_r50-d8_769x769_20k_MMOTU.py ./experiments/pspnet_r50-d8_769x769_20k_MMOTU/results/iter_80000.pth --eval mIoU

Task2: UDA semantic segmentation

 cd MMOTU_DS2Net
 
 ./tools/dist_test.sh ./experiments/DS2Net_segformerb5_769x769_40k_MMOTU/config/DS2Net_segformerb5_769x769_40k_MMOTU.py ./experiments/DS2Net_segformerb5_769x769_40k_MMOTU/results/iter_40000.pth --eval mIoU

Description of MMOTU/DS²Net

• https://arxiv.org/abs/2207.06799

If you have any question, please discuss with me by sending email to [email protected].

If you find this code useful please cite:

@article{DBLP:journals/corr/abs-2207-06799,
  author    = {Qi Zhao and
               Shuchang Lyu and
               Wenpei Bai and
               Linghan Cai and
               Binghao Liu and
               Meijing Wu and
               Xiubo Sang and
               Min Yang and
               Lijiang Chen},
  title     = {A Multi-Modality Ovarian Tumor Ultrasound Image Dataset for Unsupervised
               Cross-Domain Semantic Segmentation},
  journal   = {CoRR},
  volume    = {abs/2207.06799},
  year      = {2022},
}

References

Many thanks to their excellent works

• MMSegmentation
• MMGeneration