Created
September 25, 2019 12:19
-
-
Save mrtpk/33293e834be648a9a1d3fbffaedef54a to your computer and use it in GitHub Desktop.
Network inspired from VGG and UNET
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| # Network inspired from VGG and UNET | |
| # ref: https://towardsdatascience.com/understanding-semantic-segmentation-with-unet-6be4f42d4b47 | |
| # ref: https://tuatini.me/practical-image-segmentation-with-unet/ | |
| # ref: https://github.com/zhixuhao/unet/blob/master/model.py | |
| # import dependencies | |
| from keras.models import Model | |
| import keras.layers as layer | |
| def basic_block(input_tensor, bloc_name, n_convs, n_filters): | |
| ''' | |
| Returns basic block of convolution for the network | |
| ''' | |
| x = input_tensor | |
| name = "{}_conv{}" | |
| for i in range(0, n_convs): | |
| x = layer.Conv2D(filters=n_filters, kernel_size=(3, 3), | |
| activation='relu', padding='same', name=name.format(bloc_name, i+1))(x) | |
| # TODO: add batchnorm layer | |
| return x | |
| def get_unet(input_shape=(224, 224, 3), classes=2): | |
| ''' | |
| Returns a model inspired from VGG with UNET architecture | |
| ''' | |
| input_layer = layer.Input(input_shape, name='input_layer') | |
| # Contracting Path | |
| block1 = basic_block(input_layer, "block1", n_convs=2, n_filters=64) | |
| block1_pool = layer.MaxPool2D(pool_size=(2, 2), strides=(2,2), name="block1_pool")(block1) | |
| block2 = basic_block(block1_pool, "block2", n_convs=2, n_filters=128) | |
| block2_pool = layer.MaxPool2D(pool_size=(2, 2), strides=(2,2), name="block2_pool")(block2) | |
| block3 = basic_block(block2_pool, "block3", n_convs=3, n_filters=256) | |
| block3_pool = layer.MaxPool2D(pool_size=(2, 2), strides=(2,2), name="block3_pool")(block3) | |
| block4 = basic_block(block3_pool, "block4", n_convs=3, n_filters=512) | |
| block4_pool = layer.MaxPool2D(pool_size=(2, 2), strides=(2,2), name="block4_pool")(block4) | |
| block5 = basic_block(block4_pool, "block5", n_convs=3, n_filters=512) | |
| # block5_pool = layer.MaxPool2D(pool_size=(2, 2), strides=(2,2), name="block5_pool")(block5) | |
| # Expansive Path | |
| block6_deconv = layer.Conv2DTranspose(filters=512, kernel_size=(3,3), strides=(2, 2), | |
| padding='same', name="block6_deconv1")(block5) | |
| block6_concat = layer.concatenate([block6_deconv, block4]) | |
| block6_basic_block = basic_block(block6_concat, "block6_basic", n_convs=3, n_filters=256) | |
| block7_deconv = layer.Conv2DTranspose(filters=256, kernel_size=(3,3), strides=(2, 2), | |
| padding='same', name="block7_deconv1")(block6_basic_block) | |
| block7_concat = layer.concatenate([block7_deconv, block3]) | |
| block7_basic_block = basic_block(block7_concat, "block7_basic", n_convs=3, n_filters=128) | |
| block8_deconv = layer.Conv2DTranspose(filters=128, kernel_size=(3,3), strides=(2, 2), | |
| padding='same', name="block8_deconv1")(block7_basic_block) | |
| block8_concat = layer.concatenate([block8_deconv, block2]) | |
| block8_basic_block = basic_block(block8_concat, "block8_basic", n_convs=2, n_filters=64) | |
| block9_deconv = layer.Conv2DTranspose(filters=64, kernel_size=(3,3), strides=(2, 2), | |
| padding='same', name="block9_deconv1")(block8_basic_block) | |
| block9_concat = layer.concatenate([block9_deconv, block1]) | |
| block9_basic_block = basic_block(block9_concat, "block9_basic", n_convs=2, n_filters=64) | |
| # segmentation head | |
| block10 = basic_block(block9_basic_block, "block10", n_convs=2, n_filters=3) | |
| final_layer = layer.Conv2D(filters=classes, kernel_size=(1, 1), | |
| activation='sigmoid', padding='same', name="final_layer")(block10) | |
| return Model(inputs=input_layer, outputs=final_layer, name="vgg inspired unet") |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment