better directory structure

[wangkuiyi]

1. We need a /examples directory or python/examples.
2. Move /adversarial into /examples/fluid/advbox.
3. Move /v1_api_demo into /examples/v1

[luotao1]

@wangkuiyi @lcy-seso How about move /adversarial and v1_api_demo into models repo?

[lcy-seso]

I agree models repo need examples of Fluid so that users can learn by examples, and the examples are in one place so that users can easily find them.

[JiayiFeng]

In my opinion, advbox is a tool instead of a model. It doesn't train and produce a model. Oppositely, it takes a completed model and generates some modified images via the model.

[luotao1]

It takes a completed model and generates some modified images via the model

So it works like inference?

[JiayiFeng]

@luotao1
Indeed. However, during normal inference, we just need the value of the predicted label, while in the advbox we get the gradient of the input image and then modify the image itself until the model cannot infer it correctly.

[luotao1]

models repo doesn't contain model only. For example, in https://github.com/PaddlePaddle/models#9-image-classification, following two chapters are all tools like advbox.

• 9.1 convert Caffe model file to PaddlePaddle model file
• 9.2 convert TensorFlow model file to PaddlePaddle model file

Thus, advbox can move to models repo as well.

[wangkuiyi]

Let us move paddle:/adversarial to models:/fluid/adversarial. @gx-wind Can you please do it? Sorry that we weren't able to decide a good place to put the code initially. If we can move it, Github repo history would show you as the original author of the example.

[gx-wind]

OK. It's a good and reasonable proposal.I will move later.
Thank you all.

[luotao1]

@gx-wind You can move into https://github.com/PaddlePaddle/models/tree/develop/fluid.