CODEWITHSUNDEEP
pythontensorflowkerasconv-neural-networkloss-function
David kim
David kim

Reputation: 180

how to set appropriate input shape of model in Keras

I'm a newbie to Keras. I'm playing around Keras to get some intuition and stuck with here.

input_image = tf.keras.Input(shape=(16,16,3))
x = tf.keras.layers.Conv2D(32,(3,3), padding = 'same')(input_image)
model = tf.keras.Model(input_image , x)

model.compile(optimizer='Adam',loss = 'MSE')

inputs = np.random.normal(size = (16,16,3))
outputs = np.random.normal(size = (16,16,32))
model.fit(x = inputs , y =outputs)

I just wanted to see the output shape that model.summary says (None, 16, 16, 32). But now I have two questions. One is the output shape and another is why my code doesn't work. I hope someone tells me what I'm missing. Thanks~

Upvotes: 4

Views: 839

Answers (1)

Pygirl
Pygirl

Reputation: 13349

inputs = np.random.normal(size = (1,16,16,3)) #<---- here
outputs = np.random.normal(size = (1,16,16,32)) #<---here

They should be 4D not 3D in shape. You need to give the detail of batch also.

(batch_size, w,h,c) <---- 4D

You are missing batch_size

32,(3,3) from tf.keras.layers.Conv2D(32,(3,3), padding = 'same')(input_image)

You have 32 filters. So the channel depth will be 32. But since you have used the padding='same' so your output will have the same dimension as input. Only differ in depth.

Upvotes: 2

Related Questions