Convert ML Model using tf2onnx

Hi, I'm trying to add my ML model to the project, but I cant't figure out how to make it compatible with Lens Studio. There's a code snippet that reproduces the problem.

import tensorflow as tf

inputs = tf.keras.layers.Input(
    shape=(224, 224, 3),
    name="image"
)

x_one = tf.keras.layers.Conv2D(
    filters=16,
    kernel_size=3,
    strides=2,
    padding="same",
    activation="sigmoid"
)(inputs)

x_two = tf.keras.layers.Conv2D(
    filters=32,
    kernel_size=3,
    strides=2,
    padding="same",
    activation="sigmoid"
)(x_one)

x_three = tf.keras.layers.Conv2DTranspose(
    16,
    3,
    strides=2,
    padding="same",
    use_bias=False,
)(x_two)

x_four = tf.keras.layers.Concatenate()([x_one, x_three])

x_five = tf.keras.layers.Conv2DTranspose(
    3,
    3,
    strides=2,
    padding="same",
    use_bias=False,
)(x_four)

x_six = tf.keras.layers.Concatenate()([inputs, x_five])

model = tf.keras.Model(
    inputs=[inputs], 
    outputs=[x_six], 
    name="test-model"
)

tf.keras.models.save_model(model, "test-model")

I converted this model to ONNX using tf2onnx and the first error I got is:

The ONNX model's ConvTranspose layer `auto_pad` attribute is a DEPRECATED attribute

Then I tried to delete these attributes and after that I got an error:

ONNX model ConvTranspose layer converter only supports same padding for height and width.

Lastly, I changed the pads atrribute to 1, 1, 1, 1 but I got the error 

Shape mismatch in concat layer.

Do you have any suggestions what to do to fix these errors.