I currently have a U-net model to segment eczema regions from healthy skin. The problem I am encountering is that my model does not do a very good job in segmenting more complex shapes. Is there anything I can try to change (parameters, layers etc) to improve the performance of my model? I know preprocessing the input image can yield better results, but I feel that its current performance are a little underwhelming.
Any suggestions are appreciated. Thank you.
[400 training images with 0.1 validation split, 100 test images, input RGB image dimension is 256 x 256]
inputs = tf.keras.layers.Input((IMG_HEIGHT, IMG_WIDTH, IMG_CHANNELS))
s = tf.keras.layers.Lambda(lambda x: x / 255)(inputs)
# Contraction path
c1 = tf.keras.layers.Conv2D(16, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(s)
c1 = BatchNormalization()(c1)
c1 = tf.keras.layers.Conv2D(16, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(c1)
c1 = BatchNormalization()(c1)
#c1 = tf.keras.layers.Dropout(0.1)(c1)
p1 = tf.keras.layers.MaxPooling2D((2, 2))(c1)
c2 = tf.keras.layers.Conv2D(32, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(p1)
c2 = BatchNormalization()(c2)
c2 = tf.keras.layers.Conv2D(32, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(c2)
c2 = BatchNormalization()(c2)
#c2 = tf.keras.layers.Dropout(0.1)(c2)
p2 = tf.keras.layers.MaxPooling2D((2, 2))(c2)
c3 = tf.keras.layers.Conv2D(64, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(p2)
c3 = BatchNormalization()(c3)
c3 = tf.keras.layers.Conv2D(64, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(c3)
c3 = BatchNormalization()(c3)
#c3 = tf.keras.layers.Dropout(0.2)(c3)
p3 = tf.keras.layers.MaxPooling2D((2, 2))(c3)
c4 = tf.keras.layers.Conv2D(128, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(p3)
c4 = BatchNormalization()(c4)
c4 = tf.keras.layers.Conv2D(128, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(c4)
c4 = BatchNormalization()(c4)
#c4 = tf.keras.layers.Dropout(0.2)(c4)
p4 = tf.keras.layers.MaxPooling2D(pool_size=(2, 2))(c4)
c5 = tf.keras.layers.Conv2D(256, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(p4)
c5 = BatchNormalization()(c5)
c5 = tf.keras.layers.Conv2D(256, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(c5)
c5 = BatchNormalization()(c5)
#c5 = tf.keras.layers.Dropout(0.3)(c5)
# Expansive path
u6 = tf.keras.layers.Conv2DTranspose(128, (2, 2), strides=(2, 2), padding='same')(c5)
u6 = tf.keras.layers.concatenate([u6, c4])
c6 = tf.keras.layers.Conv2D(128, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(u6)
c6 = BatchNormalization()(c6)
c6 = tf.keras.layers.Conv2D(128, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(c6)
c6 = BatchNormalization()(c6)
#c6 = tf.keras.layers.Dropout(0.2)(c6)
u7 = tf.keras.layers.Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same')(c6)
u7 = tf.keras.layers.concatenate([u7, c3])
c7 = tf.keras.layers.Conv2D(64, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(u7)
c7 = BatchNormalization()(c7)
c7 = tf.keras.layers.Conv2D(64, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(c7)
c7 = BatchNormalization()(c7)
#c7 = tf.keras.layers.Dropout(0.2)(c7)
u8 = tf.keras.layers.Conv2DTranspose(32, (2, 2), strides=(2, 2), padding='same')(c7)
u8 = tf.keras.layers.concatenate([u8, c2])
c8 = tf.keras.layers.Conv2D(32, (3, 3), activation='relu', kernel_initializer='he_normal', padding='same')(u8)
c8 = BatchNormalization()(c8)
c8 = tf.keras.layers.Conv2D(32, (3, 3), activation='relu', kernel_initializer='he_normal', padding='same')(c8)
c8 = BatchNormalization()(c8)
#c8 = tf.keras.layers.Dropout(0.1)(c8)
u9 = tf.keras.layers.Conv2DTranspose(16, (2, 2), strides=(2, 2), padding='same')(c8)
u9 = tf.keras.layers.concatenate([u9, c1], axis=3)
c9 = tf.keras.layers.Conv2D(16, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(u9)
c9 = BatchNormalization()(c9)
c9 = tf.keras.layers.Conv2D(16, (5, 5), activation='relu', kernel_initializer='he_normal', padding='same')(c9)
c9 = BatchNormalization()(c9)
#c9 = tf.keras.layers.Dropout(0.1)(c9)
outputs = tf.keras.layers.Conv2D(1, (1, 1), activation='sigmoid')(c9)
model = tf.keras.Model(inputs=[inputs], outputs=[outputs])
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
model.summary()
results = model.fit(X_train, Y_train, validation_split=0.1, batch_size=8, epochs=28, verbose=2,steps_per_epoch=45)
x_tested = model.predict(X_test,verbose = 1)
x_tested2 = (x_tested > 0.5).astype(np.uint8)
question from:
https://stackoverflow.com/questions/66048814/increasing-image-segmentation-performance-of-deep-learning-model-u-net 
