DL augmentation package

DL augmentation package is implemented for neural network-assisted atomic electron tomography. This package has two parts; data generator codes (written in matlab) & deep learning codes (written in python, pytorch package).

For the detail, please see the paper: J. Lee, C. Jeong and Y. Yang, “Single-atom level determination of 3-dimensional surface atomic structure via neural network-assisted atomic electron tomography”, Nature Communications (2021).

Data generator for DL augmentation (matlab)

DL data generator is a package for generating input/target data for DL augmentation.

Requirements

• matlab (>= 2017a)
• GENFIRE (reconstruction algorithm)

The original GENFIRE URL is below.

(https://www.physics.ucla.edu/research/imaging/dataSoftware.html)

The GENFIRE algorithm is modified to link this DL data generator easily (GPU accelerated vesion). The modified GENFIRE alogorithm for this data generator is also uploaded and recommended.

How to start

cd ./DL_data_generator

1. Choose DLDGen_main.m for crystal structure or DLDGen_main_amorphous.m for amorphous structure.
2. Set a folder location, a file name, and some parameters for generating atomic model.
3. Run DLDGen_main.m or DLDGen_main_amorphous.m

DL augmentation (python, pytorch package)

DL augmentation takes input/target data as only mat file (in this version).

Requirements

• pytorch (>=1.1) (1.1, 1.2, 1.3, 1.4 versions were tested)
• numpy
• scipy

Reconmended environment

• Linux
• CUDA (10.1 was tested)
• GPU memory (>= 4 GB for 144³ voxel size, >= 8 GB for 256³ voxel size) (if used)
• conda environment

How to start

cd ./DL_augmentation

For training

1. Choose train_DL_augmentation_Pt.py or train_DL_augmentation_Pt.ipynb for jupyter notebook.
2. Set a folder location, a file name, and some parameters for training DL augmentation.
3. Run train_DL_augmentation_Pt.py.

For running

• Please prepare input data (its size should be equal or smaller compared to training input/target data size). Before running DL augmentation, pre-processing for input data is reconmended for matching voxels size (required) and real interpolation (optional) to reduce noise.

• Pre-processing (matlab) (If voxel size of input data is equal to training input data, this proccess can be skipped.)

  • run ./Preproccesing_tomogram_for_DL_aug/Preprocessing_for_DL_aug_input.m

• run DL_augmentation

  1. Choose run_DL_augmentation_Pt.py or run_DL_augmentation_Pt.ipynb for jupyter notebook.
  2. Set a folder location, a file name, and saved model parameter path.
  3. Run run_DL_augmentation_Pt.py.

Test for DL_augmentation

python ./DL_augmentation/run_DL_augmentation_Pt.py