用于细胞电子冷冻断层扫描分析的人工智能_AI_python_代码_下载

# One-Shot Learning With Attention-Guided Segmentation in Cryo-Electron Tomography Bo Zhou, Haisu Yu, Xiangrui Zeng, Xiaoyan Yang, Jing Zhang, and Min Xu Frontiers in Molecular Biosciences (AI in Biological and Biomedical Imaging), 2020 [[Paper](https://www.frontiersin.org/articles/10.3389/fmolb.2020.613347/full)] This repository contains the PyTorch implementation of COS-Net. ### Citation If you use this code for your research or project, please cite: @article{zhou2020one, title={One-Shot Learning With Attention-Guided Segmentation in Cryo-Electron Tomography}, author={Zhou, Bo and Yu, Haisu and Zeng, Xiangrui and Yang, Xiaoyan and Zhang, Jing and Xu, Min}, journal={Frontiers in Molecular Biosciences}, volume={7}, year={2020}, publisher={Frontiers Media SA} } ### Environment and Dependencies Requirements: * Python 3.7 * Pytorch 0.4.1 * scipy * scikit-image * opencv-python * denseCRF3D Our code has been tested with Python 3.7, Pytorch 0.4.1, CUDA 10.0 on Ubuntu 18.04. ### Dataset Setup ../ Data/ ├── subtomograms_snrINF # dataset with SNR = INF │ ├── cnt.npy # 22x1 array -- contains the number of subtomograms for each classes (22 classes in total) │ ├── data.npy # 22x1000x32x32x32 array -- contains 22 classes' subtomograms with size of 32x32x32. 1000 subtomogram storage place give, and is indexed by cnt.npy │ ├── data_seg.npy # 22x1000x32x32x32 array -- contains 22 classes' subtomogram segmentation with size of 32x32x32, corresponding with data.npy │ ├── testtom_vol_00.npy # 1xcountsx32x32x32 array -- contains 1st class's subtomograms with size of 32x32x32, counts means number of subtomogram in this class. │ ├── testtom_seg_00.npy # 1xcountsx32x32x32 array -- contains 1st class's subtomogram segmentation with size of 32x32x32, counts means number of subtomogram in this class. │ ├── testtom_vol_01.npy # please refer to above comments │ ├── testtom_seg_01.npy │ ├── testtom_vol_02.npy │ ├── testtom_seg_02.npy │ ├── testtom_vol_03.npy │ ├── testtom_seg_03.npy │ ├── testtom_vol_04.npy │ ├── testtom_seg_04.npy │ ├── testtom_vol_05.npy │ ├── testtom_seg_05.npy │ ├── testtom_vol_06.npy │ ├── testtom_seg_06.npy │ ├── testtom_vol_07.npy │ ├── testtom_seg_07.npy │ ├── testtom_vol_08.npy │ ├── testtom_seg_08.npy │ ├── testtom_vol_09.npy │ ├── testtom_seg_09.npy │ ├── testtom_vol_10.npy │ ├── testtom_seg_10.npy │ ├── testtom_vol_11.npy │ ├── testtom_seg_11.npy │ ├── testtom_vol_12.npy │ ├── testtom_seg_12.npy │ ├── testtom_vol_13.npy │ ├── testtom_seg_13.npy │ ├── testtom_vol_14.npy │ ├── testtom_seg_14.npy │ ├── testtom_vol_15.npy │ ├── testtom_seg_15.npy │ ├── testtom_vol_16.npy │ ├── testtom_seg_16.npy │ ├── testtom_vol_17.npy │ ├── testtom_seg_17.npy │ ├── testtom_vol_18.npy │ ├── testtom_seg_18.npy │ ├── testtom_vol_19.npy │ ├── testtom_seg_19.npy │ ├── testtom_vol_20.npy │ ├── testtom_seg_20.npy │ ├── testtom_vol_21.npy │ └── testtom_seg_21.npy │ ├── subtomograms_snr10000 # dataset with SNR = 10000 │ ├── cnt.npy │ └── ... └── Please refer to the above data directory setup to prepare you data, and please carefully ead the comments for the file contents. \ The one-shot training and testing process use cnt.py / data.npy / data_seg.npy files. Our dataloader will split the first 1-14 classes as training class and 15-22 classes as testing class. \ The segmentation testing process use testtom_vol_nn.npy files. ### To Run Our Code - Train and Test the one-shot model ```bash python main_oneshot.py --dataset 'snrINF' --data_root '../Data/subtomograms_snrINF/' --net 'dusescnn' --N 2 --K 1 ``` where \ `--dataset` is the dataset's name. \ `--data_root` provides the data folder directory (with structure illustrated above). \ `--net` is the network block structrue. \ `--N` is the number of way/class. \ `--K` is the number of shot/sample per class. \ Other hyperparameters can be adjusted in the code as well. - Test the model's segmentation output ```bash python main_segment.py --dataset 'snrINF' --data_root '../Data/subtomograms_snrINF/' --resume './outputs/2-way-1-shot_dusescnn_snrINF/model_25.pkl' --net 'dusescnn' --N 2 --K 1 ``` where \ `--resume` defines which checkpoint for segmentation test evaluation. \ The segmentation visualizations will be saved under './outputs/2-way-1-shot_dusescnn_snrINF/segment/'. \ The quantitative results per class will be saved in './outputs/2-way-1-shot_dusescnn_snrINF/segment/dice.txt'. \ Sample training/test scripts are provided under './scripts/' and can be directly executed. ### Contact If you have any question, please file an issue or contact the author: ``` Bo Zhou: [email protected] ```