silicon-analyser

Installation

Install from source:

pip install --upgrade .

Install from pip:

pip install --upgrade silicon-analyser

Information

Side note

Bad project name decisions might lead people to this project.

Maybe you are looking for SiliconAnalysis instead?

I'm sorry for that but I can't figure out any new project name (yet).

Acceleration

The code will use your graphic card for acceleration. (but only if correct pytorch is installed, see "Additional info" below)

Frameworks/Libraries used:

• PyQt5
• PyTorch
• Keras
• PyQtGraph

Small example

• start
• select image
• add grid
• press mouse down on image and drag your rectangle for your grid
• adjust x,y,cols,rows,width,height manualy to fit
• add label (while grid is selected)
  • give it a random name
• with that label selected, select cells for that label (for example cells that mark a "1")
• select grid (for example "grid_0" again)
• add another label (while grid is selected)
  • give it a random name
• with that label selected, select cells for that label (for example cells that mark a "0")
• with enough "1" and "0" labels drawn, click the "Compute" button
  • ai will find images in the grid that have the same properties
  • click "stop" once the results are satisfied
    • maximum for "acc" and "val_acc" is 1.00, the closer you are to those values, the better are the results
    • results depend on many factors:
      • the amount of cells you selected
      • how good your grid matches the current image
      • the quality of your image
      • ...
    • "acc" stands for "accuracy", "val" for "validation"
• found ai-cells will be drawn green

Additional info

• you might need to install cuda-specific PyTorch for accelerated computing
  • check your graphic driver version for compatible cuda version!
• Computation (currently) only happens based on active/visible grid cells (don't be fooled by accuracy of 1 just because you have only 1 label active - just activate all and use compute)

Command line

For automatically opening a file, you can pass the filepath as a filename. For example: silicon-analyser c:\my_files\image.png But keep in mind, that the program currently needs to create files in its current working directory (grid.json, rect.json).

Keys

• Use up/down/left/right to navigate
• Hold shift to move faster
• Scroll-wheel to zoom out
• Click on minimap to get directly to a position
• Right click on tree-items (left navigation menu) for additional options
• Hold down middle mouse button, to move across the screen
  • (behaviour might change in future, currently it does not behave as expected)

TODO

• undo option
• maybe use a real db in background
• some method to autofit grid
• performance improvements
• option for compute to continue from last training (currently starts fresh training)
• show loading screen on start (pytorch with cuda support takes a bit to load)
• ai-model configuration
• project management (project-file/-folder)
• possibility to rotate grid
• maybe store your model on a public place? (for others to use)