ValueError: Negative dimension size caused by subtracting 22 from 1 for 'conv3d_3/convolution' (op: 'Conv3D')

Question

I got this error message when when declaring the input layer in Keras.

Traceback (most recent call last):

File "E:/physionet/CNN_onemodel.py", line 150, in createModel model.add(Conv3D(16, (22, 5, 5), strides=(1, 2, 2), padding='valid',activation='relu',data_format= "channels_last", input_shape=input_shape))

ValueError: Negative dimension size caused by subtracting 22 from 1 for 'conv3d_3/convolution' (op: 'Conv3D') with input shapes: [?,1,22,5,3844], [22,5,5,3844,16].

Any help is appreciated.

code:

    input_shape=(1, 22, 5, 3844)
    model = Sequential()
    #C1
    model.add(Conv3D(16, (22, 5, 5), strides=(1, 2, 2), padding='valid',activation='relu',data_format= "channels_first", input_shape=input_shape))
    model.add(keras.layers.MaxPooling3D(pool_size=(1, 2, 2),data_format= "channels_first",  padding='same'))
    model.add(BatchNormalization())
    #C2
    model.add(Conv3D(32, (1, 3, 3), strides=(1, 1,1), padding='valid',data_format= "channels_first",  activation='relu'))#incertezza se togliere padding
    model.add(keras.layers.MaxPooling3D(pool_size=(1,2, 2),data_format= "channels_first", ))
    model.add(BatchNormalization())

     #C3
    model.add(Conv3D(64, (1,3, 3), strides=(1, 1,1), padding='valid',data_format= "channels_first",  activation='relu'))#incertezza se togliere padding
    model.add(keras.layers.MaxPooling3D(pool_size=(1,2, 2),data_format= "channels_first", ))
    model.add(BatchNormalization())

    model.add(Flatten())
    model.add(Dropout(0.5))
    model.add(Dense(256, activation='sigmoid'))
    model.add(Dropout(0.5))
    model.add(Dense(2, activation='softmax'))

    opt_adam = keras.optimizers.Adam(lr=0.00001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
    model.compile(loss='categorical_crossentropy', optimizer=opt_adam, metrics=['accuracy'])

bsquare bsquare · Accepted Answer · 2020-04-01T07:30:27

If you set padding = "valid"(default behavior), meaning that the automatic dimensionality reduction occurs during convolution/maxpooling and you will get negative dimensions. To make sure you get the same dimensionality after performing convolution/maxpooling as you need to set padding=same while specifying Conv3D and MaxPooling3D layers.

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten
from tensorflow.keras.layers import Conv3D, MaxPooling3D, BatchNormalization
import numpy as np

input_shape=(1, 22, 5, 3844)
model = Sequential()
    #C1
model.add(Conv3D(16, (22, 5, 5), strides=(1, 2, 2), padding='same',activation='relu',data_format= "channels_first", input_shape=input_shape))
model.add(MaxPooling3D(pool_size=(1, 2, 2),data_format= "channels_first", padding='same'))
model.add(BatchNormalization())
    #C2
model.add(Conv3D(32, (1, 3, 3), strides=(1, 1, 1), padding='same',data_format= "channels_first",  activation='relu'))#incertezza se togliere padding
model.add(MaxPooling3D(pool_size=(1, 2, 2),data_format= "channels_first", padding='same'))
model.add(BatchNormalization())
     #C3
model.add(Conv3D(64, (1, 3, 3), strides=(1, 1, 1), padding='same',data_format= "channels_first",  activation='relu'))#incertezza se togliere padding
model.add(MaxPooling3D(pool_size=(1, 2, 2), data_format= "channels_first", padding='same'))
model.add(BatchNormalization())

model.add(Flatten())
model.add(Dropout(0.5))
model.add(Dense(256, activation='sigmoid'))
model.add(Dropout(0.5))
model.add(Dense(2, activation='softmax'))

opt_adam = tf.keras.optimizers.Adam(lr=0.00001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
model.compile(loss='categorical_crossentropy', optimizer=opt_adam, metrics=['accuracy'])

print(model.summary())

Output:

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv3d (Conv3D)              (None, 16, 22, 3, 1922)   8816      
_________________________________________________________________
max_pooling3d (MaxPooling3D) (None, 16, 22, 2, 961)    0         
_________________________________________________________________
batch_normalization (BatchNo (None, 16, 22, 2, 961)    3844      
_________________________________________________________________
conv3d_1 (Conv3D)            (None, 32, 22, 2, 961)    4640      
_________________________________________________________________
max_pooling3d_1 (MaxPooling3 (None, 32, 22, 1, 481)    0         
_________________________________________________________________
batch_normalization_1 (Batch (None, 32, 22, 1, 481)    1924      
_________________________________________________________________
conv3d_2 (Conv3D)            (None, 64, 22, 1, 481)    18496     
_________________________________________________________________
max_pooling3d_2 (MaxPooling3 (None, 64, 22, 1, 241)    0         
_________________________________________________________________
batch_normalization_2 (Batch (None, 64, 22, 1, 241)    964       
_________________________________________________________________
flatten (Flatten)            (None, 339328)            0         
_________________________________________________________________
dropout (Dropout)            (None, 339328)            0         
_________________________________________________________________
dense (Dense)                (None, 256)               86868224  
_________________________________________________________________
dropout_1 (Dropout)          (None, 256)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 2)                 514       
=================================================================
Total params: 86,907,422
Trainable params: 86,904,056
Non-trainable params: 3,366
_________________________________________________________________

ValueError: Negative dimension size caused by subtracting 22 from 1 for 'conv3d_3/convolution' (op: 'Conv3D')

1 Answers