Time distributed CNNs + LSTM in Keras

cnn_lstm.py

def defModel():
    model = Sequential()
    #Izda.add(TimeDistributed(
    #    Convolution2D(40,3,3,border_mode='same'), input_shape=(sequence_lengths, 1,8,10)))
    model.add(
        TimeDistributed(
            Conv2D(32, (7, 7), padding='same', strides = 2),
            input_shape=(None, 540, 960, 2)))
    model.add(Activation('relu'))

    model.add(TimeDistributed(Conv2D(64, (5, 5), padding='same', strides = 2)))
    model.add(Activation('relu'))

    #model.add(TimeDistributed(MaxPooling2D((2,2), data_format = 'channels_first', name='pool1')))
    
    model.add(TimeDistributed(Conv2D(128, (5, 5), padding='same', strides = 2)))
    model.add(Activation('relu'))    
    
    model.add(TimeDistributed(Conv2D(128, (3, 3), padding='same')))
    model.add(Activation('relu'))
    
    model.add(TimeDistributed(Conv2D(256, (3, 3), padding='same', strides = 2)))
    model.add(Activation('relu'))
    
    model.add(TimeDistributed(Conv2D(256, (3, 3), padding='same')))
    model.add(Activation('relu'))
    
    model.add(TimeDistributed(Conv2D(256, (3, 3), padding='same', strides = 2)))
    model.add(Activation('relu'))    

    model.add(TimeDistributed(Conv2D(256, (3, 3), padding='same')))
    model.add(Activation('relu'))
    
    model.add(TimeDistributed(Conv2D(512, (3, 3), padding='same', strides = 2)))
    model.add(Activation('relu'))    
    #model.add(TimeDistributed(MaxPooling2D((2,2), data_format = 'channels_first', name='pool1')))    
    
    #model.add(TimeDistributed(Conv2D(32, (1, 1), data_format = 'channels_first')))
    #model.add(Activation('relu'))    
    
    model.add(TimeDistributed(Flatten()))
    
    #model.add(TimeDistributed(Dense(512, name="first_dense" )))
    
    #model.add(LSTM(num_classes, return_sequences=True))
    model.add(LSTM(512 , return_sequences=True))
    model.add(LSTM(512))
    model.add(Dense(128))
    model.add(Dense(3))

    model.compile(loss='mean_squared_error', optimizer='adam')  #,
    #metrics=['accuracy'])
    plot_model(model, to_file='model/model.png')
    plot_model(model, to_file='model/model_detail.png', show_shapes=True)
    return model

Can I ask how data is organized, because I'm getting an error of expecting 5 dimensions while it is getting only 4 dims

hello,
i am trying to implement CNN+LSTM, the code for the model is almost same using timedistributed layers. The model is compilng fine. I have used keras image data generators for image inputs. It gives an error on fit_generator
Error is:

ValueError: Error when checking input: expected time_distributed_1_input to have 5 dimensions, but got array with shape (4, 64, 64, 3)

I have input images of size 64x64x3, want output of 4 sample images at once. how the 5th dimension included
Any help would be appreciated
Thank you!

hello,
i am trying to implement CNN+LSTM, the code for the model is almost same using timedistributed layers. The model is compilng fine. I have used keras image data generators for image inputs. It gives an error on fit_generator
Error is:

ValueError: Error when checking input: expected time_distributed_1_input to have 5 dimensions, but got array with shape (4, 64, 64, 3)

I have input images of size 64x64x3, want output of 4 sample images at once. how the 5th dimension included
Any help would be appreciated
Thank you!

Your input shape did not match with define in the code input_shape=(None, 540, 960, 2).

I believe the input shape would be ( batch_size, time_step, rows, cols, channels. ), in which case the model will move ahead with the training process.