search

ClementPinard / FlowNetTorch

Torch implementation of Fischer et al. FlowNet training code
30 stars 6 forks source link

Hi, a problem about FlownetS.lua #5

Closed yxqlwl closed 7 years ago

yxqlwl commented 7 years ago

Hi, according to the network definition of FlowNetS in 《FlowNet: Learning Optical Flow with Convolutional Networks》，maybe conv4_1 in FlownetS.lua( and FlownetSBN.lua ) is supposed to be like the following?

 conv4_1:add(nn.SpatialConvolution(256,512,3,3,2,2,1,1))
 conv4_1:add(nn.LeakyReLU(0.1,true))
 conv4_1:add(nn.SpatialConvolution(512,512,3,3,1,1,1,1))
 conv4_1:add(nn.LeakyReLU(0.1,true))

Thanks!

ClementPinard commented 7 years ago

You are right. I changed the specifications to match the original paper. Thank you for noticing it ! However, i am not planning to redo the graphs in the README, so they may be different from the 'official' architecture.

Clément

yxqlwl commented 7 years ago

Just a small mistake:)On the whole, this code is exellent.