/PtyGer

Multi-GPU based ptychographic reconstruction tool

Primary LanguageCOtherNOASSERTION

PtyGer

A tool for ptychographic GPU(multiple)-based reconstruction.

Installation from source

export CUDAHOME=path-to-cuda
python setup.py clean
python setup.py install

Dependency

cupy - for GPU acceleration of linear algebra operations in iterative schemes. See (https://cupy.chainer.org/). For installation use

conda install -c anaconda cupy

Tests

Test PtyGer with siemens-star synthetic dataset:

cd tests/
python test.py 4 0 512 1 32 siemens4g60 -s siemens 256 60

4: number of GPUs.
0: GPU id offset (e.g., the first GPU's id is 0).
512: used for real-experiment data only. Meaningless in synthetic test run.
1: stride length of the scans.
32: number of iteration.
siemens4g60: prefix of the performance data files.
-s: accepting synthetic dataset (use -r for real-experiment dataset).
siemens: name of the dataset.
256: probe size (in this case the probe size is 256^2).
60: number of scan position (in this case the total number of scan position is 60^2=3600).

In the first running, the synthetic diffraction patterns will be generated and stored as tests/data/siemens/256data60.npy. Any repeated runnings then can directly load this data file.
The reconstructed images will be stored in tests/rec_siemens.