Add passing of args and kwargs to Model.

[hgaiser]

This allows to use args from Model (such as name).

[hgaiser]

Okay this PR grew a bit more than I thought :)

It now does:

• Pass args and kwargs to model.
• Change ResNet classes to functions returning Model (there wasn't really a reason for it to be a subclass?).
• Update the TimeDistributed version of ResNet to include the changes from https://github.com/broadinstitute/keras-resnet/pull/24 . (are these even used?)

[0x00b1]

Thanks, @hgaiser! I'll take a look today.

Update the TimeDistributed version of ResNet to include the changes from #24 . (are these even used?)

The names are used for loading weights by name.

[hgaiser]

I meant are the time distributed versions of ResNet even used :)

I know the blocks are used in keras-rcnn, but the ResNet model?

[0x00b1]

I meant are the time distributed versions of ResNet even used :)

Oops! I misread.

The TimeDistributed model is useful for video classification. It can also be used for volumetric classification (if, for some reason, you’re dealing with volumes that have an irregular number of slices). However, it’s still wonky. 😜 It’d be nice if TimeDistributed could wrap entire models so we didn’t need to provide this implementation.

[hgaiser]

Ah oke, so they do serve a purpose :) good to know

[0x00b1]

@hgaiser I want to merge this soon! Would you mind rebasing and removing the whitespace changes?

[hgaiser]

@hgaiser I want to merge this soon! Would you mind rebasing and removing the whitespace changes?

Done! The whitespace changes are intentional though, since it is no longer a class it has 'dedented' (correct word for that?).

[hgaiser]

@0x00b1 or @jhung0 can this be merged?

[hgaiser]

Can be closed since https://github.com/broadinstitute/keras-resnet/pull/27 got merged.

[0x00b1]

Thanks, @hgaiser. I was just about to ask. 😎