Questions about the network structure

[YueWuHKUST]

Hi, thanks for your great work and sharing the code.

I'm confused about the network structure. For subnetwork1, the output channel is 2, it's the complete flow of center frame right? But the input channel of subnetwork2 is 66, specified in train_refine.py. Also, the paper writes "For each pair of the consecutive frames, i frame and i+1 frame, the second subnetwork takes a sequence of estimated bidirectional flow", and said the flow is generated by first subnetwork? So where this bidirectional flow come from? Is the flow generated by FlowNet2? And you replace the flow of center frame by the result of subnetwork1?

Looking forward to your reply!