Skip to content

Latest commit

 

History

History
108 lines (80 loc) · 2.01 KB

README.md

File metadata and controls

108 lines (80 loc) · 2.01 KB

alquimodelia

Tesselo's Model Alquemy

About

Package to enable easy usage of Tesselo's common model

Features:

  • U-Net models
    • 3D and 2D architetures
    • Make variable number of layers
  • Pixel Model:
    • Make as a class
  • Models:
    • ResNet
    • LSTM
    • Outside packages
  • Test Models:
    • Working samples to test models
    • Visualize results
    • Visualize inner layers

Usage examples:


3D_UNET

from alquimodelia.unet_arch import UNet2D, UNet3D

UNet3D_model = UNet3D(
        n_filters=16,
        number_of_conv_layers=None,
        kernel_size=3,
        batchnorm=True,
        padding_style="same",
        activation_middle="relu",
        kernel_initializer="he_normal",
        timesteps=12,
        width=600,
        height=600,
        padding=None,
        num_bands=10,
        num_classes=4,
        data_format="channels_last",
)

3D_RESNET

from alquimodelia.resnet_arch import ResNet2D, ResNet3D

ResNet3D_model = ResNet3D(
        n_filters=16,
        timesteps=12,
        width=600,
        height=600,
        num_bands=10,
        num_classes=4,
        data_format="channels_last",
)

LSTM

from alquimodelia.rnn_lstm_arch import RnnLSTM

RnnLSTM_model = RnnLSTM(
        timesteps=48,
        num_bands=10,
        num_classes=12,
        activation_final="softmax",
        data_format="channels_last",
        lstm_units=(120, 80),
)

[back to usage examples]


This project is standing on the shoulders of the following giants

Harshall Lamba : https://towardsdatascience.com/understanding-semantic-segmentation-with-unet-6be4f42d4b47

Gracelyn Shi: https://towardsdatascience.com/implementing-a-resnet-model-from-scratch-971be7193718

Jason Brownlee: https://machinelearningmastery.com/keras-functional-api-deep-learning/