Negative dimension size caused by subtracting 3 from 1 for 'conv2d_2/convolution' (op: 'Conv2D') with input shapes: [?,1,74,16], [3,3,16,32]

Question

I am doing CNN image classification and got this error. Please help me to solve it. This is my code:

model = Sequential()

model.add(Conv2D(32, (3, 3), input_shape=(3, 150, 150)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2), dim_ordering="th"))

model.add(Conv2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2), dim_ordering="th"))

model.add(Conv2D(64, (3, 3))) 
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2), dim_ordering="th"))

Answer 1

You are using a very old way of specifying the order of the channels (perhaps a very old version of Keras).

"th" stands for Theano, which is deprecated by the time we speak.

Modify your code to:

model = Sequential()

model.add(Conv2D(32, (3, 3), input_shape=(150,150,3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2), dim_ordering="tf"))

model.add(Conv2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2), dim_ordering="tf"))

model.add(Conv2D(64, (3, 3))) 
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2), dim_ordering="tf"))

and your problem will be solved.

Ideally, you could set the dimension ordering globally in order to avoid specifying at each step the dimension.

Eg

import keras.backend as K
K.set_image_dim_ordering('tf')