AE-based Texture Descriptor

Autoencoder-based texture descriptor proposed in Master Thesis [TODO:cite]. The autoencoders were trained with different texture-oriented loss functions and latent space dimmensions. It can be used for the same purposes as any other descriptor.

Targets

Target defines what was the training dataset. Two targets are currently supported:

• General texture - trained with general texture dataset, as described in the thesis
• SEM texture - trained with SEM texture images

Exmple usage

Install after clonning:

pip install .

Usage:

from AEDescriptor import * 
imgage = cv.imread("image.png", cv.IMREAD_GRAYSCALE)
mode = AEDescriptor(GetModelName(loss_function=LossFunction.FFT, dim=16))
mode.Eval(image)
# => [0.67124015 0.6368097 ... 0.41859195] 16 features

To get list of all currently avaiable descriptors:

print(ListAvailableModels())