tf2show 0.0.3

tf2show prints tensorflow2's keras model pretty.

tf2show

Install

pip install tf2show

Example

import tensorflow as tf
from tf2show import tf2show

model = tf.keras.applications.ResNet50()
tf2show(model)	# show model structure
tf2show(model,"model.xlsx")    # save model structure as excel file

Description

tf2show prints tensorflow2's keras model pretty.

Below is the result of summary function provided in tensorflow2.

It's not pretty. In addition, some output has been omitted.

Model: "mobilenet_1.00_256"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         [(None, 256, 256, 3)]     0         
_________________________________________________________________
conv1_pad (ZeroPadding2D)    (None, 257, 257, 3)       0         
_________________________________________________________________
conv1 (Conv2D)               (None, 128, 128, 32)      864       
_________________________________________________________________
conv1_bn (BatchNormalization (None, 128, 128, 32)      128       
_________________________________________________________________
conv1_relu (ReLU)            (None, 128, 128, 32)      0         
_________________________________________________________________
conv_dw_1 (DepthwiseConv2D)  (None, 128, 128, 32)      288       
_________________________________________________________________
.
.
.
_________________________________________________________________
reshape_2 (Reshape)          (None, 1000)              0         
_________________________________________________________________
predictions (Activation)     (None, 1000)              0         
=================================================================
Total params: 4,253,864
Trainable params: 4,231,976
Non-trainable params: 21,888
_________________________________________________________________

Below is the output using tf2show.

It's pretty. All names are printed.

----------------------------------------------------------------------------------------------------
| LAYER                  | NAME                     | C    | H   | W    | INPUTS                   |
----------------------------------------------------------------------------------------------------
| InputLayer             | input_1                  | 3    | 256 | 256  | input_1:0                |
| ZeroPadding2D          | conv1_pad                | 3    | 257 | 257  | input_1:0                |
| Conv2D                 | conv1                    | 32   | 128 | 128  | conv1_pad                |
| BatchNormalization     | conv1_bn                 | 32   | 128 | 128  | conv1                    |
| ReLU                   | conv1_relu               | 32   | 128 | 128  | conv1_bn                 |
| DepthwiseConv2D        | conv_dw_1                | 32   | 128 | 128  | conv1_relu               |
| BatchNormalization     | conv_dw_1_bn             | 32   | 128 | 128  | conv_dw_1                |
| ReLU                   | conv_dw_1_relu           | 32   | 128 | 128  | conv_dw_1_bn             |
| Conv2D                 | conv_pw_1                | 64   | 128 | 128  | conv_dw_1_relu           |
| BatchNormalization     | conv_pw_1_bn             | 64   | 128 | 128  | conv_pw_1                |
| ReLU                   | conv_pw_1_relu           | 64   | 128 | 128  | conv_pw_1_bn             |
| ZeroPadding2D          | conv_pad_2               | 64   | 129 | 129  | conv_pw_1_relu           |
| DepthwiseConv2D        | conv_dw_2                | 64   | 64  | 64   | conv_pad_2               |
.
.
.
| Reshape                | reshape_2                |      |     | 1000 | conv_preds               |
| Activation             | predictions              |      |     | 1000 | reshape_2                |
----------------------------------------------------------------------------------------------------

It also supports saving to Excel.

This function can be useful when analyzing models.