Slim vision models should test with various dynamic shape variants

[stellaraccident]

Ideally:

• dynamic batch size
• dynamic image size

This can be done in a patchy way by modifying slim_vision_model_test.py and none-ing out various shape dims:

    input_shape = list(get_input_shape())
    input_shape[0] = None
    self.m.build(input_shape)

[allieculp]

@stellaraccident Old issue alert! Still active?

[allieculp]

@stellaraccident Old issue alert! Still active?

[allieculp]

@stellaraccident Old issue alert! Still active?

[jpienaar]

Closing as not current focus.