Need Dropout Layer in UNet and UNet++

[yudshj]

Here is an example code for the dropout layer:

def GiveMeUnet(inputImage, numFilters = 16, droupouts = 0.1, doBatchNorm = True):
    # defining encoder Path
    c1 = Conv2dBlock(inputImage, numFilters * 1, kernelSize = 3, doBatchNorm = doBatchNorm)
    p1 = tf.keras.layers.MaxPooling2D((2,2))(c1)
    p1 = tf.keras.layers.Dropout(droupouts)(p1)
    
    c2 = Conv2dBlock(p1, numFilters * 2, kernelSize = 3, doBatchNorm = doBatchNorm)
    p2 = tf.keras.layers.MaxPooling2D((2,2))(c2)
    p2 = tf.keras.layers.Dropout(droupouts)(p2)
    
    c3 = Conv2dBlock(p2, numFilters * 4, kernelSize = 3, doBatchNorm = doBatchNorm)
    p3 = tf.keras.layers.MaxPooling2D((2,2))(c3)
    p3 = tf.keras.layers.Dropout(droupouts)(p3)
    
    c4 = Conv2dBlock(p3, numFilters * 8, kernelSize = 3, doBatchNorm = doBatchNorm)
    p4 = tf.keras.layers.MaxPooling2D((2,2))(c4)
    p4 = tf.keras.layers.Dropout(droupouts)(p4)
    
    c5 = Conv2dBlock(p4, numFilters * 16, kernelSize = 3, doBatchNorm = doBatchNorm)
    
    # defining decoder path
    u6 = tf.keras.layers.Conv2DTranspose(numFilters*8, (3, 3), strides = (2, 2), padding = 'same')(c5)
    u6 = tf.keras.layers.concatenate([u6, c4])
    u6 = tf.keras.layers.Dropout(droupouts)(u6)
    c6 = Conv2dBlock(u6, numFilters * 8, kernelSize = 3, doBatchNorm = doBatchNorm)
    
    u7 = tf.keras.layers.Conv2DTranspose(numFilters*4, (3, 3), strides = (2, 2), padding = 'same')(c6)
    
    u7 = tf.keras.layers.concatenate([u7, c3])
    u7 = tf.keras.layers.Dropout(droupouts)(u7)
    c7 = Conv2dBlock(u7, numFilters * 4, kernelSize = 3, doBatchNorm = doBatchNorm)
    
    u8 = tf.keras.layers.Conv2DTranspose(numFilters*2, (3, 3), strides = (2, 2), padding = 'same')(c7)
    u8 = tf.keras.layers.concatenate([u8, c2])
    u8 = tf.keras.layers.Dropout(droupouts)(u8)
    c8 = Conv2dBlock(u8, numFilters * 2, kernelSize = 3, doBatchNorm = doBatchNorm)
    
    u9 = tf.keras.layers.Conv2DTranspose(numFilters*1, (3, 3), strides = (2, 2), padding = 'same')(c8)
    u9 = tf.keras.layers.concatenate([u9, c1])
    u9 = tf.keras.layers.Dropout(droupouts)(u9)
    c9 = Conv2dBlock(u9, numFilters * 1, kernelSize = 3, doBatchNorm = doBatchNorm)
    
    output = tf.keras.layers.Conv2D(1, (1, 1), activation = 'sigmoid')(c9)
    model = tf.keras.Model(inputs = [inputImage], outputs = [output])
    return model

[yingkaisha]

I will work on them in the coming days.