オートエンコーダの層について

Question

閲覧いただきありがとうございます。

オートエンコーダの層について、
641281
→326432
→163264
→512
→64
→512
→163264
→326432
→641281
という風に構築したいのですが、作成したモデルをmodel.sumamary()で確認したところ想定したモデル構成とは違って見えるため、質問いたしました。
間違っている箇所や思考の違う箇所等指摘していただけると幸いです。

input_img = Input(shape=(64, 128,))
hidden = Reshape((64, 128, 1), input_shape = (64, 128))(input_img)
x = Conv2D(16, (3, 3), padding='same')(hidden)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = Conv2D(32, (3, 3), padding='same')(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(64, (3, 3), padding='same')(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
encoded = MaxPooling2D((2, 2), padding='same')(x)

flatten = Dense(512, activation='relu')(encoded)
flatten = Dropout(0.2)(flatten)
flatten = Dense(64, activation='relu')(flatten)
flatten = Dropout(0.2)(flatten)
flatten = Dense(512, activation='relu')(flatten)

encoded_input = Conv2D(64, (3, 3), padding='same')(flatten)
x = BatchNormalization()(encoded_input)
x = Activation('relu')(x)
x = UpSampling2D((2, 2))(x)
x = Conv2D(32, (3, 3), padding='same')(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = UpSampling2D((2, 2))(x)
x = Conv2D(16, (3, 3), padding='same')(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = Conv2D(1, (3, 3), padding='same')(x)
x = BatchNormalization()(x)
decoded = Activation('sigmoid')(x)

autoencoder = Model(input_img, decoded)

Model: "model_7"
Layer (type) Output Shape Param #

input_8 (InputLayer) [(None, 64, 128)] 0

reshape_7 (Reshape) (None, 64, 128, 1) 0

conv2d_49 (Conv2D) (None, 64, 128, 64) 640

batch_normalization_49 (Bat (None, 64, 128, 64) 256
chNormalization)

activation_49 (Activation) (None, 64, 128, 64) 0

conv2d_50 (Conv2D) (None, 64, 128, 32) 18464

batch_normalization_50 (Bat (None, 64, 128, 32) 128
chNormalization)

activation_50 (Activation) (None, 64, 128, 32) 0

max_pooling2d_14 (MaxPoolin (None, 32, 64, 32) 0
g2D)

conv2d_51 (Conv2D) (None, 32, 64, 16) 4624

batch_normalization_51 (Bat (None, 32, 64, 16) 64
chNormalization)

activation_51 (Activation) (None, 32, 64, 16) 0

max_pooling2d_15 (MaxPoolin (None, 16, 32, 16) 0
g2D)

dense_15 (Dense) (None, 16, 32, 512) 8704

dropout_12 (Dropout) (None, 16, 32, 512) 0

dense_16 (Dense) (None, 16, 32, 64) 32832

dropout_13 (Dropout) (None, 16, 32, 64) 0

conv2d_52 (Conv2D) (None, 16, 32, 16) 9232

batch_normalization_52 (Bat (None, 16, 32, 16) 64
chNormalization)

activation_52 (Activation) (None, 16, 32, 16) 0

up_sampling2d_14 (UpSamplin (None, 32, 64, 16) 0
g2D)

conv2d_53 (Conv2D) (None, 32, 64, 32) 4640

batch_normalization_53 (Bat (None, 32, 64, 32) 128
chNormalization)

activation_53 (Activation) (None, 32, 64, 32) 0

up_sampling2d_15 (UpSamplin (None, 64, 128, 32) 0
g2D)

conv2d_54 (Conv2D) (None, 64, 128, 64) 18496

batch_normalization_54 (Bat (None, 64, 128, 64) 256
chNormalization)

activation_54 (Activation) (None, 64, 128, 64) 0

conv2d_55 (Conv2D) (None, 64, 128, 1) 577

batch_normalization_55 (Bat (None, 64, 128, 1) 4
chNormalization)

activation_55 (Activation) (None, 64, 128, 1) 0

=================================================================
Total params: 99,109
Trainable params: 98,659
Non-trainable params: 450

---

よろしくお願いいたします。