This Keras decoder maps a latent-space vector to a reconstructed image. A dense layer and reshape operation restore the encoder's pre-flattening tensor shape, and transposed-convolution layers perform the reconstruction. The decoder need not mirror the encoder exactly, but its final output shape must match the reconstruction target.

This snippet assumes that shape_before_flattening and the decoder configuration attributes have already been defined.

decoder_input = Input(shape=(self.z_dim,), name='decoder_input')

x = Dense(np.prod(shape_before_flattening))(decoder_input)
x = Reshape(shape_before_flattening)(x)

for i in range(self.n_layers_decoder):
    conv_t_layer = Conv2DTranspose(
        filters=self.decoder_conv_t_filters[i],
        kernel_size=self.decoder_conv_t_kernel_size[i],
        strides=self.decoder_conv_t_strides[i],
        padding='same',
        name='decoder_conv_t_' + str(i),
    )
    x = conv_t_layer(x)

    if i < self.n_layers_decoder - 1:
        x = LeakyReLU()(x)
        if self.use_batch_norm:
            x = BatchNormalization()(x)
        if self.use_dropout:
            x = Dropout(rate=0.25)(x)
    else:
        x = Activation('sigmoid')(x)

decoder_output = x
self.decoder = Model(decoder_input, decoder_output)

Source implementation: https://github.com/davidADSP/GDL_code/blob/master/models/AE.py