Please enable JavaScript.

Coggle requires JavaScript to display documents.

Original Code, Set counter to 0, Set counter to 0, Receive information on…

- - - - Added run identifiers and automatically saved all test settings, the trained net, and results
        
        Version 1.01
        
        Added functionality to repeatedly analyse a static set of hyperparameters
        
        Updated method of saving data as hundreds of runs were being performed
        
        Added the automatic exporting of data to excel for analysis
        
        Added functionality to update the hyperparameters
        
        Added functionality to introduce noise and artefacts to the images during pre-testing
        
        1 more item...
        
        Version 1.03
        
        Version 1.02
- - - - Save 5 KPIs, time taken, loss function, training accuracy to excel document
        
        Increase counter by 1
        
        If counter is divisible by 5
        
        Increase hyperparameter by set amount (e.g Epoch = 6)
        
        If counter is less than desired run quantity
        
        ELSE
        
        1 more item...
        
        Else
- - - - If counter is less than total runs desired
        
        Else
        
        Terminate analysis
        
        Graphs, errors, and formatting automated within Excel
- - - - Give the image a map with three dimensions.
    - - Index the image with its own map
  - - - Replace a random set of pixels between 1 and the maximum size of the image with 1
    - - Select a random X coordinate
        
        Select a random Y coordinate
        
        Create a 6x6 selection of 1 valued pixels at this location
    - - Increase each pixel in the right half of the image by the set increment
- - - - Split the database according to split ratio hyperparameter
        
        Load the pretrained network (e.g. SqueezeNet)
        
        Replace the final & learnable layer
        
        Freeze the initial layer
        
        Train the network
        
        Load the test data
        
        1 more item...
        
        Save the produced trained net
        
        Save the associated options