How is preprocessed data obtained?

[TheDarkKnight1998]

I mean your static semantic segmentation, how to remove all dynamic objects？Is it manual?

[os1a]

Hi,

No it is not manual. We train an inpainting network which is responsible to remove the dynamic objects from the image without introducing artifacts. Please check section 3.2 in the paper for details.

We train the inpainting module from:
https://github.com/JiahuiYu/generative_inpainting

But note that the inpainting module is trained on semantic maps rather than raw RGB images. This causes less artifacts and yields better results.

Best,