Architectures
We provide here a Keras model.summary() for the architectures used in the paper.
Encoder/decoder for the ASF-4 dataset
Model: "encoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 128, 56, 8)] 0
_________________________________________________________________
conv2d (Conv2D) (None, 64, 28, 32) 16416
_________________________________________________________________
batch_normalization (BatchNo (None, 64, 28, 32) 128
_________________________________________________________________
activation (Activation) (None, 64, 28, 32) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 32, 14, 64) 131136
_________________________________________________________________
batch_normalization_1 (Batch (None, 32, 14, 64) 256
_________________________________________________________________
activation_1 (Activation) (None, 32, 14, 64) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 16, 7, 128) 524416
_________________________________________________________________
batch_normalization_2 (Batch (None, 16, 7, 128) 512
_________________________________________________________________
activation_2 (Activation) (None, 16, 7, 128) 0
_________________________________________________________________
conv2d_3 (Conv2D) (None, 8, 4, 256) 2097408
_________________________________________________________________
batch_normalization_3 (Batch (None, 8, 4, 256) 1024
_________________________________________________________________
activation_3 (Activation) (None, 8, 4, 256) 0
_________________________________________________________________
flatten (Flatten) (None, 8192) 0
_________________________________________________________________
dense (Dense) (None, 3) 24579
=================================================================
Total params: 2,795,875
Trainable params: 2,794,915
Non-trainable params: 960
_________________________________________________________________
Model: "decoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_2 (InputLayer) [(None, 3)] 0
_________________________________________________________________
dense_1 (Dense) (None, 28672) 114688
_________________________________________________________________
reshape (Reshape) (None, 16, 7, 256) 0
_________________________________________________________________
activation_4 (Activation) (None, 16, 7, 256) 0
_________________________________________________________________
conv2d_transpose (Conv2DTran (None, 32, 14, 128) 2097280
_________________________________________________________________
batch_normalization_4 (Batch (None, 32, 14, 128) 512
_________________________________________________________________
activation_5 (Activation) (None, 32, 14, 128) 0
_________________________________________________________________
conv2d_transpose_1 (Conv2DTr (None, 64, 28, 64) 524352
_________________________________________________________________
batch_normalization_5 (Batch (None, 64, 28, 64) 256
_________________________________________________________________
activation_6 (Activation) (None, 64, 28, 64) 0
_________________________________________________________________
conv2d_transpose_2 (Conv2DTr (None, 128, 56, 32) 131104
_________________________________________________________________
batch_normalization_6 (Batch (None, 128, 56, 32) 128
_________________________________________________________________
activation_7 (Activation) (None, 128, 56, 32) 0
_________________________________________________________________
conv2d_transpose_3 (Conv2DTr (None, 128, 56, 8) 16392
_________________________________________________________________
activation_8 (Activation) (None, 128, 56, 8) 0
=================================================================
Total params: 2,884,712
Trainable params: 2,884,264
Non-trainable params: 448
_________________________________________________________________
Model: "autoencoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 128, 56, 8)] 0
_________________________________________________________________
encoder (Model) (None, 3) 2795875
_________________________________________________________________
decoder (Model) (None, 128, 56, 8) 2884712
=================================================================
Total params: 5,680,587
Trainable params: 5,679,179
Non-trainable params: 1,408
Encoder/decoder for the HP-10 dataset
Model: "encoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 128, 64, 20)] 0
_________________________________________________________________
conv2d (Conv2D) (None, 64, 32, 32) 40992
_________________________________________________________________
batch_normalization (BatchNo (None, 64, 32, 32) 128
_________________________________________________________________
activation (Activation) (None, 64, 32, 32) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 32, 16, 64) 131136
_________________________________________________________________
batch_normalization_1 (Batch (None, 32, 16, 64) 256
_________________________________________________________________
activation_1 (Activation) (None, 32, 16, 64) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 16, 8, 128) 524416
_________________________________________________________________
batch_normalization_2 (Batch (None, 16, 8, 128) 512
_________________________________________________________________
activation_2 (Activation) (None, 16, 8, 128) 0
_________________________________________________________________
conv2d_3 (Conv2D) (None, 8, 4, 256) 2097408
_________________________________________________________________
batch_normalization_3 (Batch (None, 8, 4, 256) 1024
_________________________________________________________________
activation_3 (Activation) (None, 8, 4, 256) 0
_________________________________________________________________
flatten (Flatten) (None, 8192) 0
_________________________________________________________________
dense (Dense) (None, 5) 40965
=================================================================
Total params: 2,836,837
Trainable params: 2,835,877
Non-trainable params: 960
_________________________________________________________________
Model: "decoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_2 (InputLayer) [(None, 5)] 0
_________________________________________________________________
dense_1 (Dense) (None, 131072) 786432
_________________________________________________________________
reshape (Reshape) (None, 32, 16, 256) 0
_________________________________________________________________
activation_4 (Activation) (None, 32, 16, 256) 0
_________________________________________________________________
conv2d_transpose (Conv2DTran (None, 64, 32, 128) 2097280
_________________________________________________________________
batch_normalization_4 (Batch (None, 64, 32, 128) 512
_________________________________________________________________
activation_5 (Activation) (None, 64, 32, 128) 0
_________________________________________________________________
conv2d_transpose_1 (Conv2DTr (None, 128, 64, 64) 524352
_________________________________________________________________
batch_normalization_5 (Batch (None, 128, 64, 64) 256
_________________________________________________________________
activation_6 (Activation) (None, 128, 64, 64) 0
_________________________________________________________________
conv2d_transpose_2 (Conv2DTr (None, 128, 64, 20) 81940
_________________________________________________________________
activation_7 (Activation) (None, 128, 64, 20) 0
=================================================================
Total params: 3,490,772
Trainable params: 3,490,388
Non-trainable params: 384
_________________________________________________________________
Model: "autoencoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 128, 64, 20)] 0
_________________________________________________________________
encoder (Model) (None, 5) 2836837
_________________________________________________________________
decoder (Model) (None, 128, 64, 20) 3490772
=================================================================
Total params: 6,327,609
Trainable params: 6,326,265
Non-trainable params: 1,344
Encoder/decoder for the LPD-5 dataset
Model: "encoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_3 (InputLayer) [(None, 192, 112, 10)] 0
_________________________________________________________________
conv2d_4 (Conv2D) (None, 96, 56, 32) 20512
_________________________________________________________________
batch_normalization_6 (Batch (None, 96, 56, 32) 128
_________________________________________________________________
activation_8 (Activation) (None, 96, 56, 32) 0
_________________________________________________________________
conv2d_5 (Conv2D) (None, 48, 28, 64) 131136
_________________________________________________________________
batch_normalization_7 (Batch (None, 48, 28, 64) 256
_________________________________________________________________
activation_9 (Activation) (None, 48, 28, 64) 0
_________________________________________________________________
conv2d_6 (Conv2D) (None, 24, 14, 128) 524416
_________________________________________________________________
batch_normalization_8 (Batch (None, 24, 14, 128) 512
_________________________________________________________________
activation_10 (Activation) (None, 24, 14, 128) 0
_________________________________________________________________
conv2d_7 (Conv2D) (None, 12, 7, 256) 2097408
_________________________________________________________________
batch_normalization_9 (Batch (None, 12, 7, 256) 1024
_________________________________________________________________
activation_11 (Activation) (None, 12, 7, 256) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 21504) 0
_________________________________________________________________
dense_2 (Dense) (None, 5) 107525
=================================================================
Total params: 2,882,917
Trainable params: 2,881,957
Non-trainable params: 960
_________________________________________________________________
Model: "decoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_4 (InputLayer) [(None, 5)] 0
_________________________________________________________________
dense_3 (Dense) (None, 344064) 2064384
_________________________________________________________________
reshape_1 (Reshape) (None, 48, 28, 256) 0
_________________________________________________________________
activation_12 (Activation) (None, 48, 28, 256) 0
_________________________________________________________________
conv2d_transpose_3 (Conv2DTr (None, 96, 56, 128) 2097280
_________________________________________________________________
batch_normalization_10 (Batc (None, 96, 56, 128) 512
_________________________________________________________________
activation_13 (Activation) (None, 96, 56, 128) 0
_________________________________________________________________
conv2d_transpose_4 (Conv2DTr (None, 192, 112, 64) 524352
_________________________________________________________________
batch_normalization_11 (Batc (None, 192, 112, 64) 256
_________________________________________________________________
activation_14 (Activation) (None, 192, 112, 64) 0
_________________________________________________________________
conv2d_transpose_5 (Conv2DTr (None, 192, 112, 10) 40970
_________________________________________________________________
activation_15 (Activation) (None, 192, 112, 10) 0
=================================================================
Total params: 4,727,754
Trainable params: 4,727,370
Non-trainable params: 384
_________________________________________________________________
Model: "autoencoder"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_3 (InputLayer) [(None, 192, 112, 10)] 0
_________________________________________________________________
encoder (Model) (None, 5) 2882917
_________________________________________________________________
decoder (Model) (None, 192, 112, 10) 4727754
=================================================================
Total params: 7,610,671
Trainable params: 7,609,327
Non-trainable params: 1,344