Same for GlobalMaxPooling2D
keras from master, tensorflow from tf-nightly-gpu
```def disc():
def conv(i, deph, size, std, usebn = True):
i = Conv2D(filters = deph, kernel_size = size, strides = 1, padding = 'same')(i)
if usebn:
i = bn(i)
i = relu(i)
i = MaxPooling2D(std)(i)
return i
inp = Input(shape = d_shape)
i = inp
i = conv(i, depth, size, 2, False)
i = conv(i, depth, size, 2)
i = conv(i, depth, size, 2)
i = conv(i, depth, size, 2)
i = GlobalMaxPooling2D(i)
i = Dense(len(labels), activation = 'softmax')(i)
m = Model(inp, i)
GlobalMaxPooling2D()(i)
Damm, pretty obvious, thank you.
Most helpful comment
GlobalMaxPooling2D()(i)