Add all necessary files into my own project folder

recognition/s4557453_ISIC-Improved_UNet/modules.py

@@ -0,0 +1,158 @@
+from keras.layers import BatchNormalization, Conv2D, UpSampling2D, LeakyReLU, Dropout, Add, Concatenate
+from keras import Input, Model
+
+
+def contextModule(inputs, filters):
+    """
+    Implemented based on: https://arxiv.org/abs/1802.10508v1
+    Two 3x3x3 convolutional layers and a dropout layer (p = 0.3) in between.
+    Use leaky ReLU nonlinearities with a negative slope of 10^−2
+    """
+
+    conv1 = Conv2D(filters, (3, 3), strides=1, padding="same")(inputs)
+    leak1 = LeakyReLU(0.01)(conv1)
+    norm1 = BatchNormalization()(leak1)
+
+    drop = Dropout(0.3)(norm1)
+
+    conv2 = Conv2D(filters, (3, 3), strides=1, padding="same")(drop)
+    leak2 = LeakyReLU(0.01)(conv2)
+    norm2 = BatchNormalization()(leak2)
+
+    return norm2
+
+
+def localizationModule(inputs, filters):
+    """
+    Implemented based on: https://arxiv.org/abs/1802.10508v1
+    Consists of a 3x3x3 convolution followed by a 1x1x1 convolution that halves the number of feature maps.
+    """
+
+    conv1 = Conv2D(filters, (3, 3), strides=1, padding="same")(inputs)
+    leak1 = LeakyReLU(0.01)(conv1)
+    norm1 = BatchNormalization()(leak1)
+
+    conv2 = Conv2D(filters/2, (1, 1), strides=1, padding="same")(norm1)
+    leak2 = LeakyReLU(0.01)(conv2)
+    norm2 = BatchNormalization()(leak2)
+
+    return norm2
+
+
+def upsamplingModule(inputs, filters):
+    """
+    Implemented based on: https://arxiv.org/abs/1802.10508v1
+    upsampling the low resolution feature maps, which is done by means of a simple upscale that repeats the
+    feature voxels twice in each spatial dimension, followed by a 3x3x3 convolution that halves the number
+    of feature maps.
+    """
+    upsample = UpSampling2D((2, 2))(inputs)
+    conv1 = Conv2D(filters, (3, 3), strides=1, padding="same")(upsample)
+    leak1 = LeakyReLU(0.01)(conv1)
+    norm1 = BatchNormalization()(leak1)
+
+    return norm1
+
+
+def getModel(input_size):
+    """
+    Implement the Improved UNet as per: https://arxiv.org/abs/1802.10508v1
+    """
+
+    # Entry block
+    inputs = Input(shape=input_size)
+
+    # Layer 1 (16 filter) - Context Pathway
+    conv1 = Conv2D(16, (3, 3), strides=1, padding="same")(inputs)
+    leak1 = LeakyReLU(0.01)(conv1)
+
+    cont1 = contextModule(leak1, 16)
+
+    add1 = Add()([leak1, cont1])
+
+    # Layer 2 (32 filter) - Context Pathway
+    conv2 = Conv2D(32, (3, 3), strides=2, padding="same")(add1)
+    leak2 = LeakyReLU(0.01)(conv2)
+
+    cont2 = contextModule(leak2, 32)
+
+    add2 = Add()([leak2, cont2])
+
+    # Layer 3 (64 filter) - Context Pathway
+    conv3 = Conv2D(64, (3, 3), strides=2, padding="same")(add2)
+    leak3 = LeakyReLU(0.01)(conv3)
+
+    cont3 = contextModule(leak3, 64)
+
+    add3 = Add()([leak3, cont3])
+
+    # Layer 4 (128 filter) - Context Pathway
+    conv4 = Conv2D(128, (3, 3), strides=2, padding="same")(add3)
+    leak4 = LeakyReLU(0.01)(conv4)
+
+    cont4 = contextModule(leak4, 128)
+
+    add4 = Add()([leak4, cont4])
+
+    # Layer 5 (256 filter) - Bottleneck
+    conv5 = Conv2D(256, (3, 3), strides=2, padding="same")(add4)
+    leak5 = LeakyReLU(0.01)(conv5)
+
+    cont5 = contextModule(leak5, 256)
+
+    add5 = Add()([leak5, cont5])
+
+    # Layer 5 (128 filter) - Bottleneck
+    upsamp1 = upsamplingModule(add5, 128)
+
+    concat1 = Concatenate()([upsamp1, add4])
+
+    # Layer 4 (128 and 64 filters) - Localization Pathway
+    local1 = localizationModule(concat1, 128)
+    upsamp2 = upsamplingModule(local1, 64)
+
+    concat2 = Concatenate()([upsamp2, add3])
+
+    # Layer 3 (64 and 32 filters) - Localization Pathway
+    local2 = localizationModule(concat2, 64)
+    upsamp3 = upsamplingModule(local2, 32)
+
+    concat3 = Concatenate()([upsamp3, add2])
+
+    seg1 = Conv2D(2, (3, 3), strides=1, padding="same")(local2)
+    seg1 = LeakyReLU(0.01)(seg1)
+
+    # Layer 2 (32 and 16 filters) - Localization Pathway
+    local3 = localizationModule(concat3, 32)
+    upsamp4 = upsamplingModule(local3, 16)
+
+    concat4 = Concatenate()([upsamp4, add1])
+
+    seg2 = Conv2D(2, (3, 3), strides=1, padding="same")(local3)
+    seg2 = LeakyReLU(0.01)(seg2)
+
+    segUp1 = UpSampling2D((2, 2))(seg1)
+    add6 = Add()([segUp1, seg2])
+
+    # Layer 1 (32 filter) - Localization Pathway
+    conv6 = Conv2D(32, (3, 3), strides=1, padding="same")(concat4)
+
+    seg3 = Conv2D(2, (3, 3), strides=1, padding="same")(conv6)
+    seg3 = LeakyReLU(0.01)(seg3)
+
+    segUp2 = UpSampling2D((2, 2))(seg2)
+    add7 = Add()([segUp2, seg3])
+
+    output = Conv2D(2, (3, 3), strides=1, padding="same", activation="softmax")(add7)
+    model = Model(inputs, output)
+
+    return model
+
+
+def main():
+    model = get_model((256, 256, 1))
+    print(model.summary())
+
+
+if __name__ == "__main__":
+    main()