why didn't you use the depth-wise separable convolution?

[Seojii]

Hi, Thank you for your great work! I have two questions about your paper!

1. As I wrote the title, I would like to know why you didn't use depth-wise separable convolution in you backbone. Since you said that you got idea from the Xception, I think you probably considered it.

2. What would you expect if the model is trained with depth-wise separable conv?

Thanks.

[MrForExample]

They probably using some hyperparameter search technique to find out they structure, using because it worked can't tell much reason in that

[xavysp]

It's a great question, sorry I did not answer you in the right time. First, my personal objective about the work is make it simpler. And I used the classical CNN. Maybe a deep study, using different type of CNN, will be great. I have faced, trying to reach the objective, different problems and I started solving them.