neonsecret/pytorchgui

Graphical User Interface for Pytorch

Python backend for graph-based processing, designed for flow-based/node-based visual scripting editors. It is the backbone of the Ryven project, but it can be used for other applications as well.

If you are not already familiar with flow-based visual scripting and are looking for a specification, see here.

Installation

pip install pytorchgui

or from sources:

git clone https://github.com/leon-thomm/pytorchgui
cd pytorchgui
pip install .

Dependencies

None! pytorchgui runs completely on standard python modules, no additional libraries required, which makes it very compatible.

Usage

Using pytorchgui directly to run projects made with pytorchgui-based editors, the following code example gives some intuition about the process:

import pytorchgui as ptgui
import json
import sys

if __name__ == '__main__':
    # get a working project file path
    if len(sys.argv) < 2:
        sys.exit('please provide a project file path')
    fpath = sys.argv[1]
    try:
        f = open(fpath)
        f.close()
    except FileNotFoundError:
        sys.exit(f'could not open file {fpath}')

    # read project file
    with open(fpath, 'r') as f:
        project: dict = json.loads(f.read())

    # run pytorchgui
    session = ptgui.Session()
    session.load(project)

    # and now we can manually access all components, for example:
    scripts = session.scripts
    flow1 = scripts[0].flow
    my_node = flow1.nodes[-1]
    my_node.update()

You can also use it for other purposes. A mostly auto-generated documentation is available here.

Main Features

• load & save into and from JSON-compatible dictionaries
• variables system with update mechanism to build nodes that automatically adapt to change of data
• built in logging based on python's logging module
• powerful nodes system which lets you do anything, simple and unrestricted
• dynamic nodes registration mechanism to register and unregister nodes at runtime
• actions system for nodes
• data and exec flow support - unlike lots of other solutions out there, pytorchgui supports exec flows

For a more detailed overview, see the docs.