Unknow error

[eftekhar-hossain]

Hello
When I add the ChannelAttention2D layer it trains correctly. However when I try to load the model it through an exception that -

Unknown layer: ChannelAttention2D.

If you provide a full keras notebook then it will be very helpful for us.

[vinayak19th]

Yeah the issue would be custom object loading in tensorflow. You'll have to load the model like this

from visual_attention import PixelAttention2D , ChannelAttention2D
loaded_model = tf.keras.models.load_model('/tmp/model',custom_objects={"ChannelAttention2D":ChannelAttention2D})

Lemme know if that fixes the issue

[rubenfonnegra]

Dear Vinayak
I'm having a similar issue when trying to load a saved model. I'm loading it as follows
custom_obj = {"PixelAttention2D":PixelAttention2D}
model = tensorflow.keras.load_model('model.h5', custom_objects = custom_obj)

However, I'm obtaining the following error:
File "/home/user/venvs/venv/lib/python3.8/site-packages/keras/engine/base_layer.py", line 784, in from_config
return cls(**config)
TypeError: __init__() missing 1 required positional argument: 'nf'
I think it's related to the input parameter requiered to use the PixelAttention2D layer, despite I haven't figured out how to fix it. Hope you could help.
Regards!

[mustofahisyam]

Dear Vinayak I'm having a similar issue when trying to load a saved model. I'm loading it as follows custom_obj = {"PixelAttention2D":PixelAttention2D} model = tensorflow.keras.load_model('model.h5', custom_objects = custom_obj)

However, I'm obtaining the following error: File "/home/user/venvs/venv/lib/python3.8/site-packages/keras/engine/base_layer.py", line 784, in from_config return cls(**config) TypeError: __init__() missing 1 required positional argument: 'nf' I think it's related to the input parameter requiered to use the PixelAttention2D layer, despite I haven't figured out how to fix it. Hope you could help. Regards!

I'm having the same issue. What i did was modify the init function and give the nf argument a default value which is the value of the CNN filter number i used.
def __init__(self, nf=32, **kwargs):

and comment or remove this line:
# config.update({"Att_filters": self.nf})

And it worked.