How modify depht of model 16x16x4 to 16x16x10

[viennois]

Hello
I would like to use the ten bands of Sentinel 2 at 10m resolution and apply the "createmodel1.py" script provided for 4 spectral bands at 10m. I have 16x16x10 patches as sinput but I don't know how modify the model in the script?
I am a beginner in phyton and CNN.
Thanks for the help
Gaëlle

[remicres]

Hi Gaëlle, it is quite straightforward.

You just have to change the number of components per pixel of the input placeholder:
Before:

x = tf.compat.v1.placeholder(tf.float32, [None, None, None, 4], name="x")

After:

x = tf.compat.v1.placeholder(tf.float32, [None, None, None, 10], name="x")

As you can see that is still the TF1 API, and you can follow the tutorial using TensorflowModelTrain with the generated SavedModel.

Soon we will provide pure python examples that employ the TF2 API (with Keras)!

[viennois]

Thank you, it works !