explaination regarding FlowNet2

[dagnichz]

i'm having a hard time understanding why you need FlowNet2 when performing video_inpaining.py
to my understanding , FlowNet2 provides optical flow for the unmasked frames.
However, in your paper you did not used Flownet2 but created DFC-net to reconstruct flow in masked regions.
do I understand correctly? if so , can you explain why the Flownet2 model is needed in the inference code ?

thanks in advance