This repository contains the Local Derivative Patterns as a Deep Learning layer implemented in the paper. The example on LFW - Labeled Faces in the Wild dataset compares a model with LDP layers and a baseline CNN model.

model_ldp - test accuracies for example runs:

          0.9342 0.9473, 0.9429, 0.9298, 0.8991, 0.9078, 0.9342, 0.9254, 0.9122, 0.9166 

model_base - test accuracies for example runs:

          0.9122, 0.9035, 0.9035, 0.8947, 0.9035, 0.9078, 0.9210, 0.8991, 0.9078, 0.9122

x1 = LDP(mode='single', alpha='0')(x1)
x2 = LDP(mode='single',alpha='45')(x2)
x3 = LDP(mode='single',alpha='90')(x3)
x4 = LDP(mode='single',alpha='135')(x4)

x = LDP(mode='mean')(x)

x = LDP(mode='multi')(x)

Reference study:

Akgun, Devrim. "TensorFlow based deep learning layer for Local Derivative Patterns." Software Impacts 14 (2022): 100452 https://doi.org/10.1016/j.simpa.2022.100452

Similar studies:

https://github.com/dwday/lbp_tensorflow_python

https://github.com/dwday/lbp_pytorch_python-