Project with system which can reduce image dimensionality using linear neural network.

It uses numpy for matrix multiplication. Gradient descent algorithm, forward propagation and dataset loader are created without usage of Deep Learning algorithms. Supports only CPU computing.

To download project:

git clone https://github.com/Vadbeg/autoencoder.git

To install all libraries you need, print in autoencoder directory:

pip install -r requirements.txt

It will install all essential libraries

After libraries installation you need to adjust configs. Config is located in config.py file. Config example:

class Config:

    image_path = 'test_images/test2.jpg'

    image_size = (256, 256)
    slide_window = (16, 16)

    num_of_hidden_neurons = 64
    learning_rate = 0.001
    adaptive_lr = False

    min_error: float = 0.03

    n_epochs = 150

Now you can train network and perform tests:

• Training phase you can find in start_trainig.py script.
• Script with tests and plots you can find in build_plots.py.

• numpy - The math framework used.

• Vadim Titko aka Vadbeg - LinkedIn | GitHub