Error loading pretrained model in `keras 3.9.2`

[markveillette]

When loading the pretrained model from huggingface using keras 3.9.2 creates an error.

# Load pretrained model
cnn = keras.models.load_model(model_file,compile=False

generates

TypeError: <class 'keras.src.models.functional.Functional'> could not be deserialized properly. Please ensure that components that are Python object instances (layers, models, etc.) returned by get_config()are explicitly deserialized in the model'sfrom_config() method.

Appears to have trouble loading the custom layer CoordConv2D

[markveillette]

Three ways to deal with this (will fix the notebooks and test soon)

Option 1: Pass custom_objects to load_model

Modify the call to load_model as follows:

from tornet.models.keras.layers import CoordConv2D
cnn = keras.models.load_model(model_file,compile=False,custom_objects={'CoordConv2D':CoordConv2D})

Option 2: Load weights from prebuilt model

from tornet.models.keras.cnn_baseline import build_model
cnn = build_model(start_filters=48)
cnn.load_weights(model_file)

Option 3: Downgrade to keras 3.6