You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I'm training a dataset using greyscale data normalized between 0 and 1 with a shape 1024,256,1. Following the documentation the input is converted to 3 layers using
BACKBONE = 'resnet18'
base_model = sm.Unet(backbone_name=BACKBONE, encoder_weights='imagenet',
classes=1, activation='sigmoid')
inp = Input(shape=(None, None, 1))
l1 = Conv2D(3, (1, 1))(inp) # map N channels data to 3 channels
out = base_model(l1)
model = Model(inp, out, name=base_model.name)
With the Backbone set to resnet, training is successful. However when I try another backbone such as efficientnet I get the following error
model, history = main(model, train_gen, val_gen, callbacks, epochs = epochs) # next section explains the use of sys.exit
File "/.../...py", line 247, in main
history = model.fit(train_gen, epochs=epochs, validation_data=val_gen,
File "/.../lib/python3.9/site-packages/keras/utils/traceback_utils.py", line 67, in error_handler
raise e.with_traceback(filtered_tb) from None
File "/.../...py", line 97, in __getitem__
x = self.preprocessing(x)
File "/.../lib/python3.9/site-packages/segmentation_models/__init__.py", line 43, in wrapper
return func(*args, **new_kwargs)
File "/.../lib/python3.9/site-packages/segmentation_models/__init__.py", line 34, in wrapper
return func(*args, **kwargs)
File "/.../lib/python3.9/site-packages/classification_models/models_factory.py", line 78, in wrapper
return func(*args, **new_kwargs)
File "/.../lib/python3.9/site-packages/classification_models/models/senet.py", line 435, in preprocess_input
return imagenet_utils.preprocess_input(x, mode='torch', **kwargs)
IndexError: index 1 is out of bounds for axis 3 with size 1
The text was updated successfully, but these errors were encountered:
I'm training a dataset using greyscale data normalized between 0 and 1 with a shape 1024,256,1. Following the documentation the input is converted to 3 layers using
With the Backbone set to resnet, training is successful. However when I try another backbone such as efficientnet I get the following error
The text was updated successfully, but these errors were encountered: