TODO

retry submission 19 but train on ALL patients in train

train one binary network

train a network that should predict ellipse center + size (similar to YOLO)

train a network that, given the ellipse center and the image, tries to generate a human-looking mask

inception module
==================
def Inception5(x, name):
    """
    Inception-v2 module as described in Figure 5. of
    "Revisiting the inception architecture for computer vision"
    """
    x1 = Convolution2D(8, 1, 1, border_mode='same', activation='relu')(x)
    x1 = Convolution2D(16, 3, 3, border_mode='same', activation='relu')(x1)
    x1 = Convolution2D(16, 3, 3, border_mode='same', activation='relu', subsample=(2, 2))(x1)
    
    x2 = Convolution2D(8, 1, 1, border_mode='same', activation='relu')(x)
    x2 = Convolution2D(16, 3, 3, border_mode='same', activation='relu', subsample=(2, 2))(x2)
    
    x3 = MaxPooling2D((2, 2))(x)
    x3 = Convolution2D(16, 1, 1, border_mode='same', activation='relu')(x3)
    
    x4 = Convolution2D(16, 1, 1, border_mode='same', subsample=(2, 2))(x)
    
    x = merge([x1, x2, x3, x4], mode='concat', concat_axis=1, name=name)
    return x