Andrey Rykov

@lunardog thanks for your comments. Actually, you're totally right, `prior_boxes` depend on the net architecture and don't depend on the dataset. Unfortunately, I cannot make my dataset public because of...

@Iflier surely, you need to select validation data according to your problem. I've never claimed that the example is the only correct way to do it, in my problem it...

@KevinYuk the authors use [this model](https://gist.github.com/weiliu89/2ed6e13bfd5b57cf81d6). I'm not sure, but I suppose that it is pre-trained on ImageNet, but you'd better check.

@hzycn you need to rename `get_output_shape` to `compute_output_shape` in custom layers. However there might be other issues, as the implementation was not tested with `Keras 2` api.

@hzycn is it a missprint in your comment? You need `compute_output_shape`, not `computer_output_shape`

@BenMacKenzie not exactly. You need to rename ```python def get_output_shape_for(self, input_shape): ... ``` to ```python def compute_output_shape(self, input_shape): ... ``` It's just a `Keras` 2.0 change in API.

@nsknsl I believe, you need to read [this part](https://keras.io/getting-started/functional-api-guide/) of `Keras` docs.

I have no time to review now. Moreover, there are some redundant commits in this PR. Keras 2.0 support is not a hard thing to do, there is only a...

@chenwgen could you, please, be more specific about what part of the code are you concerned?

@oarriaga Actually, these auxiliary zeros are needed for `Keras`. It cannot handle `y_true` and `y_pred` of different sizes, so, even if in `y_true` there is no need to provide priors...