add regnet, res2net, resnest, sknet encoders

[azkalot1]

Integrating some encoders using https://github.com/rwightman/pytorch-image-models

[azkalot1]

Build fails since batchnorm in some of the encoders requires >= 2 samples in the batch.
I suggest test can be changes from

def get_sample(model_class):
    if model_class in [smp.Unet, smp.Linknet, smp.FPN, smp.PSPNet, smp.UnetPlusPlus]:
        sample = torch.ones([1, 3, 64, 64])
    elif model_class == smp.PAN:
        sample = torch.ones([2, 3, 256, 256])
    elif model_class == smp.DeepLabV3:
        sample = torch.ones([2, 3, 128, 128])
    else:
        raise ValueError("Not supported model class {}".format(model_class))
    return sample

to simply

def get_sample(model_class):
    if model_class in [smp.Unet, smp.Linknet, smp.FPN, smp.PSPNet, smp.UnetPlusPlus,  smp.PAN,  smp.DeepLabV3]:
        sample = torch.ones([2, 3, 256, 256])
    else:
        raise ValueError("Not supported model class {}".format(model_class))
    return sample

[qubvel]

It is better to keep sample as small as possible, because it will increase test time significantly (only cpu is available). May be it is better to add one more 'if' case.

[qubvel]

Or set batchnorms to eval mode :)

[azkalot1]

Should I just use model.eval()?

[JulienMaille]

@azkalot1 May I ask you for feedback on the potential caveats of #373 ? Thanks