kwargs not supported

[bonlime]

This issue occurs for all models in the repo but i'll explain it in case of Unet
If you try to pass any additional arguments to Unet model, they get into **kwargs and are passed to keras_applications.get_submodules_from_kwargs function

segmentation_models/segmentation_models/models/unet.py

Line 211 in 6fecc72

backend, layers, models, keras_utils = get_submodules_from_kwargs(kwargs)

which DOESN'T allow any extra arguments! and raises TypeError
https://github.com/keras-team/keras-applications/blob/71acdcd98088501247f4b514b7cbbdf8182a05a4/keras_applications/__init__.py#L13-L21
it makes the following line useless:

segmentation_models/segmentation_models/models/unet.py

Line 226 in 6fecc72

**kwargs,

I think the easiest way to fix this issue is to write a custom get_submodules_from_kwargs function. I can make PR if you find it helpfull

[qubvel]

Hi @bonlime
What is the case you want to pass additional kwargs?
I guess only mobilenet support additional arguments

[qubvel]

I would be glad to fix it with your PR 👍

[bonlime]

I'm passing input_tensor argument. all backbone models accept it

[JonnoFTW]

I'm using the latest code from github and get an error later on about backend being NoneType. I got it working with:

from keras import layers, utils, models
from segmentation_models.models import unet
model = unet.Unet('efficientnetb3', classes=num_classes, activation='softmax',
                     backend=K, layers=layers, utils=utils, models=models)

[qubvel]

@JonnoFTW Do not import Unet from .models, look at example in readme