TFGENZOO.layers.resnet module¶

TFGENZOO.layers.resnet.ShallowConnectedResNet(inputs: tensorflow.python.keras.engine.input_layer.Input, cond: tensorflow.python.keras.engine.input_layer.Input = None, width: int = 512, out_scale: int = 2, connect_type: str = 'whole')[source]¶

ResNet of OpenAI’s Glow with Connection

Parameters

inputs (tf.Tensor) – input tensor rank == 4
cond (tf.Tensor) – input tensor rank == 4 (optional)
width (int) – hidden width
out_scale (int) – output channel width scale

Returns

tf.keras.Model

Return type

model

Sources:

https://github.com/openai/glow/blob/master/model.py#L420-L426

Examples

>>> inputs = tf.keras.Input([16, 16, 2])
>>> cond = None
>>> sr = ShallowConnectedResNet(inputs)
>>> sr.summary()
Model: "model"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to
==================================================================================================
input_1 (InputLayer)            [(None, 16, 16, 2)]  0
__________________________________________________________________________________________________
conv2d_8 (Conv2D)               (None, 16, 16, 512)  10241       input_1[0][0]
__________________________________________________________________________________________________
tf_op_layer_Relu_8 (TensorFlowO [(None, 16, 16, 512) 0           conv2d_8[0][0]
__________________________________________________________________________________________________
conv2d_9 (Conv2D)               (None, 16, 16, 512)  2360321     tf_op_layer_Relu_8[0][0]
__________________________________________________________________________________________________
tf_op_layer_Relu_9 (TensorFlowO [(None, 16, 16, 512) 0           conv2d_9[0][0]
__________________________________________________________________________________________________
tf_op_layer_concat (TensorFlowO [(None, 16, 16, 514) 0           tf_op_layer_Relu_9[0][0]
                                                                input_1[0][0]
__________________________________________________________________________________________________
conv2d_zeros_4 (Conv2DZeros)    (None, 16, 16, 4)    18512       tf_op_layer_concat[0][0]
==================================================================================================
Total params: 2,389,074
Trainable params: 2,389,072
Non-trainable params: 2
__________________________________________________________________________________________________
>>> cond = tf.keras.Input([16, 16, 128])
>>> sr = ShallowResNet(inputs, cond, connect_type="cond")
>>> sr.summary()
sr.summary()
Model: "model_4"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to
==================================================================================================
input_1 (InputLayer)            [(None, 16, 16, 2)]  0
__________________________________________________________________________________________________
input_2 (InputLayer)            [(None, 16, 16, 128) 0
__________________________________________________________________________________________________
tf_op_layer_concat_3 (TensorFlo [(None, 16, 16, 130) 0           input_1[0][0]
                                                                input_2[0][0]
__________________________________________________________________________________________________
conv2d_14 (Conv2D)              (None, 16, 16, 512)  600065      tf_op_layer_concat_3[0][0]
__________________________________________________________________________________________________
tf_op_layer_Relu_14 (TensorFlow [(None, 16, 16, 512) 0           conv2d_14[0][0]
__________________________________________________________________________________________________
conv2d_15 (Conv2D)              (None, 16, 16, 512)  2360321     tf_op_layer_Relu_14[0][0]
__________________________________________________________________________________________________
tf_op_layer_Relu_15 (TensorFlow [(None, 16, 16, 512) 0           conv2d_15[0][0]
__________________________________________________________________________________________________
tf_op_layer_concat_4 (TensorFlo [(None, 16, 16, 640) 0           tf_op_layer_Relu_15[0][0]
                                                                input_2[0][0]
__________________________________________________________________________________________________
conv2d_zeros_7 (Conv2DZeros)    (None, 16, 16, 4)    23048       tf_op_layer_concat_4[0][0]
==================================================================================================
Total params: 2,983,434
Trainable params: 2,983,432
Non-trainable params: 2
__________________________________________________________________________________________________

TFGENZOO.layers.resnet.ShallowConnectedResNetlikeSPADE(inputs: tensorflow.python.keras.engine.input_layer.Input, cond: tensorflow.python.keras.engine.input_layer.Input, width: int = 512, out_scale: int = 2, connect_type: str = 'whole')[source]¶

ResNet of OpenAI’s Glow with Connection .. note:: WIP now…

Parameters

inputs (tf.Tensor) – input tensor rank == 4
cond (tf.Tensor) – input tensor rank == 4 (optional)
width (int) – hidden width
out_scale (int) – output channel width scale

Returns

tf.keras.Model

Return type

model

Sources:

https://github.com/openai/glow/blob/master/model.py#L420-L426 https://github.com/NVlabs/SPADE

Examples:

TFGENZOO.layers.resnet.ShallowResNet(inputs: tensorflow.python.keras.engine.input_layer.Input, cond: tensorflow.python.keras.engine.input_layer.Input = None, width: int = 512, out_scale: int = 2)[source]¶

ResNet of OpenAI’s Glow

Parameters

inputs (tf.Tensor) – input tensor rank == 4
cond (tf.Tensor) – input tensor rank == 4 (optional)
width (int) – hidden width
out_scale (int) – output channel width scale

Returns

tf.keras.Model

Return type

model

Sources:

https://github.com/openai/glow/blob/master/model.py#L420-L426

Note

This layer is not Residual Network because this layer does not have Skip connection

Examples

>>> inputs = tf.keras.Input([16, 16, 2])
>>> cond = None
>>> sr = ShallowResNet(inputs)
>>> sr.summary()
Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
input_1 (InputLayer)         [(None, 16, 16, 2)]       0
_________________________________________________________________
conv2d (Conv2D)              (None, 16, 16, 512)       10241
_________________________________________________________________
tf_op_layer_Relu (TensorFlow [(None, 16, 16, 512)]     0
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 16, 16, 512)       2360321
_________________________________________________________________
tf_op_layer_Relu_1 (TensorFl [(None, 16, 16, 512)]     0
_________________________________________________________________
conv2d_zeros (Conv2DZeros)   (None, 16, 16, 4)         18440
=================================================================
Total params: 2,389,002
Trainable params: 2,389,000
Non-trainable params: 2
_________________________________________________________________
>>> cond = tf.keras.Input([16, 16, 128])
>>> sr = ShallowResNet(inputs, cond)
>>> sr.summary()
Model: "model_1"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to
==================================================================================================
input_1 (InputLayer)            [(None, 16, 16, 2)]  0
__________________________________________________________________________________________________
input_3 (InputLayer)            [(None, 16, 16, 128) 0
__________________________________________________________________________________________________
tf_op_layer_concat_1 (TensorFlo [(None, 16, 16, 130) 0           input_1[0][0]
                                                                input_3[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D)               (None, 16, 16, 512)  600065      tf_op_layer_concat_1[0][0]
__________________________________________________________________________________________________
tf_op_layer_Relu_2 (TensorFlowO [(None, 16, 16, 512) 0           conv2d_2[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D)               (None, 16, 16, 512)  2360321     tf_op_layer_Relu_2[0][0]
__________________________________________________________________________________________________
tf_op_layer_Relu_3 (TensorFlowO [(None, 16, 16, 512) 0           conv2d_3[0][0]
__________________________________________________________________________________________________
conv2d_zeros_1 (Conv2DZeros)    (None, 16, 16, 4)    18440       tf_op_layer_Relu_3[0][0]
==================================================================================================
Total params: 2,978,826
Trainable params: 2,978,824
Non-trainable params: 2
__________________________________________________________________________________________________