Deep-Flow-Guided-Video-Inpainting icon indicating copy to clipboard operation
Deep-Flow-Guided-Video-Inpainting copied to clipboard

Flow inpainting along the motion boundary

Open gaochen315 opened this issue 5 years ago • 3 comments

First of all, thanks for the great work and making the code available.

I have a question regarding the flow inpainting stage. The provided DFC weight is mainly for inpainting the flow of the foreground object, which is relatively easier because there is no motion boundary issue. However, if the hole has an arbitrary shape, then we do need to pay attention to the motion boundary. And I believe this motivates the key contribution of this paper, which is the coarse-to-fine flow completion. Thus, I was wondering if you plan to release the weights for the all subnetworks, so that we can all benefit from it.

Also, could you please provide some additional guidance on training the DFC-NET? Again, thanks for your great work. Looking forward to hearing from you.

gaochen315 avatar Jul 25 '19 18:07 gaochen315

Hi @gaochen315 , thanks for your interest in our work. Firstly, we will give some additional guidance in the tools/README later. For the refine stage, we do not have any recent plan for releasing the weight.

nbei avatar Jul 26 '19 02:07 nbei

Hi @gaochen315 , the full model for DAVIS have been updated in our git repo. The full model for YTB will also be updated recently after quick check and clean. More important, the code for MS inference has been modified for fixing bugs and saving GPU memory. Please update your local repo.

nbei avatar Aug 06 '19 01:08 nbei

Hi @nbei,

Thank you so much for releasing the weights. Looking forward to the amazing results!

gaochen315 avatar Aug 06 '19 02:08 gaochen315