AIDeveloper 0.0.7

• Import assistant to allow loading of non-squared (and oddly shaped) images
• Export assistant to convert images back to .png or .jpeg accessible from the interactive confusion matrices and on the main menu Edit->Export selected to .png/.jpg
• Allow user defined loss, learning rate, and optimizer. You can even change of of those in-between epochs (see "Expert Mode")
• Rename conda_list.txt to aid_dependencies.txt
• Properly reference aid_dependencies_OS.txt to solve #4
• Rename frontend.py to aid_frontend.py

AIDeveloper 0.0.6

Changelog:
• Support for Mac, Linux and Windows
• Support non-quadratic images
• Added an update checker: Help->Check for updates
• few bug fixes

AIDeveloper 0.0.5

Changelog:
• Enormously speeded up all augmentation routines by leveraging opencv
• Implemented Blurring options: Average, Gaussian and Motion blurring
• Set a Minimum and Maximum for “doubleSpinBox_GaussianNoiseMean” (now, Gaussian Noise mean can be set to negative values in main window and also the fitting popup)
• QRegExpValidator for dropout-"lineEdit" to avoid invalid user-input
• Cleaner organization of GUI on "Define Model" - Box on "Build" - Tab
• Enormously speeded up "Data Overview" Box, which used to cause a slow response of the main screen when many files were loaded and clicked
• Added Icons for "Normalization", "Nr. Epochs" and "Model architecture"
• In "Assess Model"-Tab -> "Classify unlabelled data" there are now 2 options: either all Scores and Predictions are saved to separate excel file, OR: the predictions are added to the .rtdc file (You can then open the file in ShapeOut and for example plot prediction vs. another feature)
• Moreover, scores of up to 9 classes can be exported to userdef1...userdef9
• More Metrics optionally available during training: Precision, Recall, F1. Those are also saved to meta.xlsx. Those additional metrics can be selected on "Expert"-Tab
• To allow addition of more metrics, “.compile” was removed from each model in model_zoo.py. Compilation is now always done by AIDeveloper, considering user-specific metrics. AIDeveloper always compiles using: loss='categorical_crossentropy' and optimizer='adam'. (So far this used to be a good choice for all image classification jobs. I you need freedom to change these parameters, let me know ([email protected] or on GitHub -> Issues).
• Drag'n drop table on Build-Tab : When clicking the header (at T,V or shuffle), all items are checked/unchecked
• Added SplashScreen (image showing at Program start-up)
• Print in console which device is used when AID is started (useful when using AID from script together with NVIDIA GPU)

AIDeveloper 0.0.5_dev2

Allow images with odd dimensions

AIDeveloper 0.0.5_dev1

Changelog:
• Removed a bug concerning the Confusion matrix (there were issues when the first class was not zero, or classes were missing)
• To support non-squared images, the function gen_crop_img in aid_img.py was updated to allow for simultaneous cropping and padding to result in squared images of user specified dimension

AIDeveloper 0.0.4

Initial release