Displot is a pre-trained machine learning driven semiconductor surface analysis program. It allows for automated detection of threading dislocations on images of surfaces obtained using ECCI. The underlying neural network is based on the FusionNet architecture.
16 GB of RAM or more is recommended to avoid OOM errors. If a GPU is available, it should be automatically detected and used during prediction.
This program is built using Python 3.7. If you are going to run it from a GNU/Linux based OS, you will likely already have it installed. If you are going to run it from a Windows based machine, it is recommended that you download an Anaconda distribution.
First clone the repository using git:
$ git clone https://github.com/bjstarosta/displot
Or manually download and unpack the source code into a directory within Python's PATH.
The required environment can be reproduced using Anaconda/Miniconda. Depending on whether the OS you are installing on is GNU/Linux or Windows based, use the appropriate file when creating the environment:
$ conda create --name displot --file env-linux64.txt
OR
$ conda create --name displot --file env-windows.txt
Once the environment is replicated, activate it using:
$ conda activate displot
Finally, due to GitHub filesize limits, you will need to separately download the latest neural network model (about 225MB) and place it in the displot/weights directory. This will be available for public download soon.
The user interface can be started by running the following command from the software directory:
$ python -m displot
Distributed under the GNU GPLv3 License. See LICENSE for more information.
We use SemVer for versioning. For the versions available, see the tags on this repository.
- E-mail: bohdan.starosta@strath.ac.uk