Added annotation typing to mobilenet

[frgfm]

Hi there!

As per #2025, annotation typing are welcome in torchvision. So, this PR focuses on torchvision.models.mobilenet this PR!

Any feedback is welcome!

[pmeier]

Only incremental changes needed from the other PRs. Otherwise LGTM.

[frgfm]

Only incremental changes needed from the other PRs. Otherwise LGTM.

Fixed! Although, I'm still getting an issue I didn't manage to get around:

torchvision/models/mobilenet.py:82: error: List item 1 has incompatible type
"Conv2d"; expected "ConvBNReLU"  [list-item]
                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
                ^
torchvision/models/mobilenet.py:83: error: List item 2 has incompatible type
Module; expected "ConvBNReLU"  [list-item]
                norm_layer(oup),
                ^
torchvision/models/mobilenet.py:154: error: Argument 1 to "append" of "list"
has incompatible type Module; expected "ConvBNReLU"  [arg-type]
                    features.append(block(input_channel, output_channel, s...

[fmassa]

@pmeier any way of telling mypy that ConvBNReLU inherits from Module?

[pmeier]

I think annotating layers: List[nn.Module] = [] should be sufficient. If the annotation is left out mypy assumes the first type that is added to the list is the only one. Thus, it probably assumes that layers is List[ConvBNReLU] and thus complains.

@frgfm You can check which type mypy uses for a variable by simply inserting reveal_type(layers) into the code and run mypy over it.

[frgfm]

@frgfm You can check which type mypy uses for a variable by simply inserting reveal_type(layers) into the code and run mypy over it.

Neat trick @pmeier 👌 Your suggestion is working, just pushed it!

[fmassa]

Thanks!