[GitHub] timprepscius opened a new issue #12221: mxnet keras padding strides, maybe not correct?

[GitBox <gi...@apache.org>]

timprepscius opened a new issue #12221: mxnet keras padding strides, maybe not correct?
URL: https://github.com/apache/incubator-mxnet/issues/12221
 
 
   ```
   import keras
   from keras.models import Sequential, Model
   from keras.layers import Dense, Dropout, Flatten, LSTM, ConvLSTM2D, Activation, Reshape, Input, Concatenate, concatenate
   from keras.layers import Conv2D, MaxPooling2D, AveragePooling2D, ZeroPadding2D
   from keras.layers.normalization import BatchNormalization
   from keras import backend as K
   
   i = Input(shape=(1, 256, 256))
   c0 = Conv2D(7, kernel_size=3, activation="relu", padding="same", strides=1)
   m0 = MaxPooling2D(pool_size=3, padding="same", strides=2)
   c1 = Conv2D(7, kernel_size=7, activation="relu", padding="same", strides=2)
   
   
   l = i
   l = c0(l)
   l = m0(l)
   l = c1(l)
   
   m = Model(inputs=[i], outputs=[l])
   
   m.compile(
       loss=keras.losses.mean_squared_error,
       optimizer=keras.optimizers.Adadelta()
   )
   
   print(m.summary())
   ```
   
   results in:
   ```
   _________________________________________________________________
   Layer (type)                 Output Shape              Param #   
   =================================================================
   input_1 (InputLayer)         (None, 1, 256, 256)       0         
   _________________________________________________________________
   conv2d_1 (Conv2D)            (None, 7, 256, 256)       70        
   _________________________________________________________________
   max_pooling2d_1 (MaxPooling2 (None, 7, 128, 128)       0         
   _________________________________________________________________
   conv2d_2 (Conv2D)            (None, 7, 64, 64)         2408      
   =================================================================
   Total params: 2,478
   Trainable params: 2,478
   Non-trainable params: 0
   _________________________________________________________________
   None
   ```
   
   (The output sized changes when stride is not 1, even though padding was "same")
   
   I think it should result in:
   ```
   _________________________________________________________________
   Layer (type)                 Output Shape              Param #   
   =================================================================
   input_1 (InputLayer)         (None, 1, 256, 256)       0         
   _________________________________________________________________
   conv2d_1 (Conv2D)            (None, 7, 256, 256)       70        
   _________________________________________________________________
   max_pooling2d_1 (MaxPooling2 (None, 7, 256, 256)       0         
   _________________________________________________________________
   conv2d_2 (Conv2D)            (None, 7, 256, 256)       2408      
   =================================================================
   Total params: 2,478
   Trainable params: 2,478
   Non-trainable params: 0
   _________________________________________________________________
   None
   
   ```
   
   But I could be wrong of course.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services